Large language models (LLMs) are increasingly used in scientific domains. While they can produce reasoning-like content via methods such as chain-of-thought prompting, these outputs are typically unstructured and informal, obscuring whether models truly understand the fundamental reasoning paradigms that underpin scientific inference. To address this, we introduce a novel task named Latent Reasoning Chain Extraction (ARCHE), in which models must decompose complex reasoning arguments into combinations of standard reasoning paradigms in the form of a Reasoning Logic Tree (RLT). In an RLT, all reasoning steps are explicitly categorized as one of three variants of Peirce’s fundamental inference modes: deduction, induction, or abduction. To facilitate this task, we release ARCHE Bench, a new benchmark derived from 70 Nature Communications articles, including more than 1,900 references and 38,000 viewpoints. We propose two logic-aware evaluation metrics: Entity Coverage (EC) for content completeness and Reasoning Edge Accuracy (REA) for step-by-step logical validity. Evaluations on 10 leading LLMs on ARCHE Bench reveal that models exhibit a trade-off between REA and EC, and none are yet able to extract a complete and standard reasoning chain. These findings highlight a substantial gap between the abilities of current reasoning models and the rigor required for scientific argumentation.

Current Multimodal Chain-of-Thought (MCoT) methods suffer from low-quality multimodal reasoning, characterized by overthinking on simple queries and inefficient utilization of visual information, resulting in vast inefficient and ineffective computations. In this paper, we discover that Multimodal Large Language Models (MLLMs) possess inherent capabilities to distinguish between simple and difficult queries and enhance task-related visual information, which remain underutilized by existing approaches. Based on this insight, we propose Self-Driven Refined Multimodal CoT (SDR-MCoT), a training-free framework that mitigates these issues through two self-driven modules. First, our selective thinking module employs entropy-based confidence estimation to determine whether queries require detailed reasoning, preventing overthinking on simple questions. Second, our step-wise visual enhancement module strengthens attention to relevant visual regions at each reasoning step without inserting additional tokens, achieving fine-grained visual grounding and enhancement with minimal overhead. Moreover, SDR-MCoT can be seamlessly integrated into various MLLMs, offering a practical solution for improving multimodal reasoning. Comprehensive experiments across eight benchmarks from diverse domains (multimodal reasoning, visual understanding, hallucination, and mathematical reasoning) demonstrate that SDR-MCoT consistently outperforms existing MCoT methods on four different base models with reduced overhead. For instance, on Qwen2-VL-7B, our method improves average accuracy by over 6% while reducing token consumption by approximately 60% compared to zero-shot CoT.

Shengji Tang, Jianjian Cao, Weihao Lin, Jiale Hong, Bo Zhang, Shuyue Hu, Lei Bai, Tao Chen, Wanli Ouyang, Peng Ye. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Yiqun Zhang, Peng Ye, Xiaocui Yang, Shi Feng, Shufei Zhang, Lei Bai, Wanli Ouyang, Shuyue Hu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Abstract Existing image restoration methods primarily rely on the posterior distribution of natural images but are often limited by their dependence on known degradations and supervised training. To this end, we propose Generative Diffusion Prior (GDP), an unsupervised sampling-based framework that effectively models posterior distributions for image and video restoration. GDP utilizes a single pre-trained denoising diffusion probabilistic model (DDPM) to solve a wide range of linear, non-linear, and blind inverse problems without explicit degradation assumptions. Specifically, GDP systematically explores a conditional guidance protocol, which proves more practical and effective than conventional methods of adding guidance. Furthermore, GDP incorporates a degradation model optimization mechanism during the denoising process, enabling blind image restoration. Besides, we introduce a patch-based strategy, allowing GDP to handle images of arbitrary resolution. We extensively evaluate GDP on multiple image and video restoration tasks, including super-resolution, deblurring, inpainting, and colorization, as well as more challenging applications such as low-light enhancement, HDR recovery, and LDR video enhancement. Experimental results demonstrate that GDP outperforms leading unsupervised methods across diverse benchmarks in both reconstruction accuracy and perceptual quality, while demonstrating robust generalization to images and videos of any size. Our project page at https://generativediffusionprior.github.io/.

Neural Architecture Search (NAS) has attracted increasing attention in recent years because of its capability to design neural networks automatically. Among them, differential NAS approaches such as DARTS, have gained popularity for search efficiency. However, they still suffer from three main issues, that are, the weak stability due to the performance collapse, the poor generalization ability of the searched architectures, and the inferior robustness to different kinds of proxies (i.e., computationally reduced search configurations). To solve the search stability and searched architecture's generalization problems, a simple-but-effective regularization method, termed as Beta-Decay, is proposed to regularize the DARTS-based NAS searching process (referred as $\beta$β-DARTS). Specifically, Beta-Decay regularization can impose constraints to keep the value and variance of activated architecture parameters from being too large, thereby ensuring fair competition among architecture parameters and making the supernet less sensitive to the impact of input on the operation set. In-depth theoretical analyses on how it works and why it works are provided, and comprehensive experiments on a variety of search spaces and datasets validate that Beta-Decay regularization can help to stabilize the searching process and make the searched network more transferable across different datasets. To address the proxy robustness problem, we first benchmark differentiable NAS methods under a wide range of proxy data, proxy channels, proxy layers, and proxy epochs, since the robustness of NAS under different kinds of proxies has not been explored before. We then conclude some interesting findings and find that $\beta$β-DARTS always achieves the best result among all compared NAS methods under almost all proxy settings. We further introduce the novel flooding regularization to the weight optimization of $\beta$β-DARTS (termed as Bi-level regularization), and experimentally and theoretically verify its effectiveness for improving the proxy robustness of differentiable NAS.

Generating lifelike human motions from descriptive texts has experienced remarkable research focus in recent years, propelled by the emerging requirements of digital humans. Despite impressive advances, existing approaches are often constrained by limited control modalities, task specificity, and focus solely on body motion representations. In this paper, we present MotionGPT-2, a unified Large Motion-Language Model (LMLM) that addresses these limitations. MotionGPT-2 accommodates multiple motion-relevant tasks and supports multimodal control conditions through pre-trained Large Language Models (LLMs). It quantizes multimodal inputs—such as text and single-frame poses—into discrete, LLM-interpretable tokens, seamlessly integrating them into the LLM’s vocabulary. These tokens are then organized into unified prompts, guiding the LLM to generate motion outputs through a pretraining-then-finetuning paradigm. We also show that the proposed MotionGPT-2 is highly adaptable to the challenging 3D holistic motion generation task, enabled by the innovative motion discretization framework, Part-Aware VQVAE, which facilitates fine-grained representations of body and hand movements. Extensive experiments and visualizations validate the effectiveness of our method, demonstrating the adaptability of MotionGPT-2 across motion generation, motion captioning, and generalized motion completion tasks.

Precipitation nowcasting based on radar echoes plays a crucial role in monitoring extreme weather and supporting disaster prevention. Although deep learning approaches have achieved significant progress, they still face notable limitations. For example, deterministic models tend to produce over-smoothed predictions, which struggle to capture extreme events and fine-scale precipitation patterns. Probabilistic generative models, due to their inherent randomness, often show fluctuating performance across different metrics and rarely achieve consistently optimal results. Furthermore, precipitation nowcasting is typically evaluated using multiple metrics, some of which are inherently conflicting. For instance, there is often a trade-off between the Critical Success Index (CSI) and the False Alarm Ratio (FAR), making it challenging for existing models to deliver forecasts that perform well on both metrics simultaneously. To address these challenges, we introduce preference optimization into precipitation nowcasting for the first time, motivated by the success of reinforcement learning from human feedback in large language models. Specifically, we propose SynCast, which lever-ages the two-stage post-training framework of Diffusion Sequential Preference Optimization (Diffusion-SPO) to progressively align conflicting metrics. In the first stage, the framework focuses on reducing FAR to deliver clean and high-precision predictions. Building on this foundation, the second stage further optimizes CSI under strict FAR constraints, thereby achieving synergistic improvements across these conflicting metrics. Experiments on three radar precipitation datasets demonstrate that SynCast reduces FAR while improving CSI, and achieves performance comparable to state-of-the-art methods. Furthermore, we verify that the post-training framework of Diffusion-SPO is compatible with multiple diffusion models for precipitation nowcasting, demonstrating its generalizability. The code for SynCast is available at https://github.com/Dtdtxuky/SynCast.

Transformer-based models have significantly advanced image captioning through self-attention mechanisms and parallel computation. However, existing methods typically adopt teacher-forcing strategies during training by conditioning the decoder exclusively on ground-truth tokens, whereas at inference, captions are generated autoregressively based solely on previously predicted tokens. Such discrepancy between training and inference conditions leads to a progressive accumulation of prediction errors, resulting in captions that deviate significantly from the visual content. While scheduled sampling strategies mitigate this issue, directly integrating them disrupts Transformer parallelism and overlooks visual saliency differences. To tackle these limitations, we present VMFormer, a multi-stage decoding Transformer featuring a Visual-Aware Scheduled Sampling (VASS) module that bridges this gap through two key innovations: 1) A two-stage decoding scheme where an initial self-study stage generates candidate tokens, followed by a hybrid stage dynamically blending ground-truth references with predictions via a visual clues controlled gate. 2) A cognitive-inspired prioritization mechanism that retains visual keywords (nouns/verbs/attributes) in early training phases before transitioning to linguistic refinements, mirroring human captioning patterns. Crucially, the VASS module preserves the parallel computational strengths of Transformer architectures and is designed as a plug-and-play component, readily adaptable to Transformer-based captioning models. Experiments on challenging MSCOCO achieve 142.2 CIDEr. We extend VMFormer to video captioning and demonstrate consistent improvements on MSRVTT and MSVD datasets.

Large Language Models (LLMs) have achieved remarkable success and have been applied across various scientific fields, including chemistry. However, many chemical tasks require the processing of visual information, which cannot be successfully handled by existing chemical LLMs. This brings a growing need for models capable of integrating multimodal information in the chemical domain. In this paper, we introduce ChemVLM, an open-source chemical multimodal large language model specifically designed for chemical applications. ChemVLM is trained on a carefully curated bilingual multimodal dataset that enhances its ability to understand both textual and visual chemical information, including molecular structures, reactions, and chemistry examination questions. We develop three datasets for comprehensive evaluation, tailored to Chemical Optical Character Recognition (OCR), Multimodal Chemical Reasoning (MMCR), and Multimodal Molecule Understanding tasks. We benchmark ChemVLM against a range of open-source and proprietary multimodal large language models on various tasks. Experimental results demonstrate that ChemVLM achieves competitive performance across all evaluated tasks.

3D Gaussian Splatting (3DGS) has shown promising performance in novel view synthesis. Previous methods adapt it to obtaining surfaces of either individual 3D objects or within limited scenes. In this paper, we make the first attempt to tackle the challenging task of large-scale scene surface reconstruction. This task is particularly difficult due to the high GPU memory consumption, different levels of details for geometric representation, and noticeable inconsistencies in appearance. To this end, we propose GigaGS, the first work for high-quality surface reconstruction for large-scale scenes using 3DGS. GigaGS first applies a partitioning strategy based on the mutual visibility of spatial regions, which effectively grouping cameras for parallel processing. To enhance the quality of the surface, we also propose novel multi-view photometric and geometric consistency constraints based on Level-of-Detail representation. In doing so, our method can reconstruct detailed surface structures. Comprehensive experiments are conducted on various datasets. The consistent improvement demonstrates the superiority of GigaGS.

Weed control is a critical challenge in modern agriculture, as weeds compete with crops for essential nutrient resources, significantly reducing crop yield and quality. Traditional weed control methods, including chemical and mechanical approaches, have real-life limitations such as associated environmental impact and efficiency. An emerging yet effective approach is laser weeding, which uses a laser beam as the stem cutter. Although there have been studies that use deep learning in weed recognition, its application in intelligent laser weeding still requires a comprehensive understanding. Thus, this study serves the first empirical study on weed recognition for laser weeding. To increase the efficiency of laser beam cut and avoid damaging the crops of interest, the laser beam shall be directly aimed at the weed root. Yet, weed stem detection remains an under-explored problem. We integrate the detection of crop and weed with the localization of weed stem into one end-to-end system. To train and validate the proposed system in a real-life scenario, we curate and construct a high-quality weed stem detection dataset with human annotations. The dataset consists of 7,161 high-resolution pictures collected in the field with annotations of 11,151 instances of weed. The dataset will be released upon acceptance. Experimental results show that, in contrast to seminal weed recognition systems, the proposed system can efficiently improve the weeding accuracy by 5.05% and reduce the energy cost by 32.3%.

Elucidating the functional mechanisms of the primary visual cortex (V1) remains a fundamental challenge in systems neuroscience. Current computational models face two critical limitations, namely the challenge of cross-modal integration between partial neural recordings and complex visual stimuli, and the inherent variability in neural characteristics across individuals, including differences in neuron populations and firing patterns. To address these challenges, we present a multi-modal identifiable variational autoencoder (miVAE) that employs a two-level disentanglement strategy to map neural activity and visual stimuli into a unified latent space. This framework enables robust identification of cross-modal correlations through refined latent space modeling. We complement this with a novel score-based attribution analysis that traces latent variables back to their origins in the source data space. Evaluation on a large-scale mouse V1 dataset demonstrates that our method achieves state-of-the-art performance in cross-individual latent representation and alignment, without requiring subject-specific fine-tuning, and exhibits improved performance with increasing data size. Significantly, our attribution algorithm successfully identifies distinct neuronal subpopulations characterized by unique temporal patterns and stimulus discrimination properties, while simultaneously revealing stimulus regions that show specific sensitivity to edge features and luminance variations. This scalable framework offers promising applications not only for advancing V1 research but also for broader investigations in neuroscience.

Haoyang Su, Renqi Chen, Shixiang Tang, Zhenfei Yin, Xinzhe Zheng, Jinzhe Li, Biqing Qi, Qi Wu, Hui Li, Wanli Ouyang, Philip Torr, Bowen Zhou, Nanqing Dong. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Jiakang Yuan, Xiangchao Yan, Bo Zhang, Tao Chen, Botian Shi, Wanli Ouyang, Yu Qiao, Lei Bai, Bowen Zhou. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Human’s perception of the visual world is shaped by the stereo processing of 3D information. Understanding how the brain perceives and processes 3D visual stimuli in the real world has been a longstanding endeavor in neuroscience. Towards this goal, we introduce a new neuroscience task: decoding 3D visual perception from EEG signals, a neuroimaging technique that enables real-time monitoring of neural dynamics enriched with complex visual cues. To provide the essential benchmark, we first present EEG-3D, a pioneering dataset featuring multimodal analysis data and extensive EEG recordings from 12 subjects viewing 72 categories of 3D objects rendered in both videos and images. Furthermore, we propose Neuro-3D, a 3D visual decoding framework based on EEG signals. This framework adaptively integrates EEG features derived from static and dynamic stimuli to learn complementary and robust neural representations, which are subsequently utilized to recover both the shape and color of 3D objects through the proposed diffusion-based colored point cloud decoder. To the best of our knowledge, we are the first to explore EEG-based 3D visual decoding. Experiments indicate that Neuro-3D not only reconstructs colored 3D objects with high fidelity, but also learns effective neural representations that enable insightful brain region analysis. The code and dataset are available at https://github.com/gzq17/neuro-3D.

Traditional spatiotemporal models generally rely on task-specific architectures, which limit their generalizability and scalability across diverse tasks due to domain-specific design requirements. In this paper, we introduce UniSTD, a unified Transformer-based framework for spatiotemporal modeling, which is inspired by advances in recent foundation models with the two-stage pretraining-then-adaption paradigm. Specifically, our work demonstrates that task-agnostic pretraining on 2D vision and vision-text datasets can build a generalizable model foundation for spatiotemporal learning, followed by specialized joint training on spatiotemporal datasets to enhance task-specific adaptability. To improve the learning capabilities across domains, our framework employs a rank-adaptive mixture-of-expert adaptation by using fractional interpolation to relax the discrete variables so that can be optimized in the continuous space. Additionally, we introduce a temporal module to incorporate temporal dynamics explicitly. We evaluate our approach on a large-scale dataset covering 10 tasks across 4 disciplines, demonstrating that a unified spatiotemporal model can achieve scalable, cross-task learning and support up to 10 tasks simultaneously within one model while reducing training costs in multi-domain applications. Code will be available at https://github.com/1hunters/UniSTD.

Accurate acquisition of surface meteorological conditions at arbitrary locations holds significant importance for weather forecasting and climate simulation. Meteorological states derived from satellite observations are often provided in the form of low-resolution grid fields. If spatial interpolation is applied directly to obtain meteorological states for specific locations, there will often be significant discrepancies compared to actual observations. Existing downscaling methods for acquiring meteorological state information at higher resolutions commonly overlook the correlation with satellite observations. To bridge the gap, we propose Satellite-observations Guided Diffusion Model (SGD), a conditional diffusion model pre-trained on ERA5 reanalysis data with satellite observations (GridSat) as conditions, which is employed for sampling downscaled meteorological states through a zero-shot guided sampling strategy and patch-based methods. During the training process, we propose to fuse the information from GridSat satellite observations into ERA5 maps via the attention mechanism, enabling SGD to generate atmospheric states that align more accurately with actual conditions. In the sampling, we employed optimizable convolutional kernels to simulate the upscale process, thereby generating high-resolution ERA5 maps using low-resolution ERA5 maps as well as observations from weather stations as guidance. Moreover, our devised patch-based method promotes SGD to generate meteorological states at arbitrary resolutions. Experiments demonstrate SGD fulfills accurate meteorological states downscaling to 6.25km. The code is available at https://github.com/Tusiwei/SGD

Much previous AI research has focused on developing monolithic models to maximize their intelligence, with the primary goal of enhancing performance on specific tasks. In contrast, this work attempts to study using LLM-based agents to design collaborative AI systems autonomously. To explore this problem, we first introduce ComfyBench to evaluate agents’s ability to design collaborative AI systems in ComfyUI. ComfyBench is a comprehensive benchmark comprising 200 diverse tasks covering various instruction-following generation challenges, along with detailed annotations for 3,205 nodes and 20 workflows. Based on ComfyBench, we further develop ComfyAgent, a novel framework that empowers LLM-based agents to autonomously design collaborative AI systems by generating workflows. ComfyAgent is based on two core concepts. First, it represents workflows with code, which can be reversibly converted into workflows and executed as collaborative systems by the interpreter. Second, it constructs a multi-agent system that cooperates to learn from existing workflows and generate new workflows for a given task. While experimental results demonstrate that ComfyAgent achieves a comparable resolve rate to o1-preview and significantly surpasses other agents on ComfyBench, ComfyAgent has resolved only 15% of creative tasks. LLM-based agents still have a long way to go in autonomously designing collaborative AI systems. Progress with ComfyBench is paving the way for more intelligent and autonomous collaborative AI systems. Our code is available at: https://github.com/xxyQwQ/ComfyBench.

Vision-language models (VLMs) have shown remarkable advancements in multimodal reasoning tasks. However, they still often generate inaccurate or irrelevant responses due to issues like hallucinated image understandings or unrefined reasoning paths. To address these challenges, we introduce Critic-V, a novel framework inspired by the Actor-Critic paradigm to boost the reasoning capability of VLMs. This framework decouples the reasoning process and critic process by integrating two independent components: the Reasoner, which generates reasoning paths based on visual and textual inputs, and the Critic, which provides constructive critique to refine these paths. In this approach, the Reasoner generates reasoning responses according to text prompts, which can evolve iteratively as a policy based on feedback from the Critic. This interaction process was theoretically driven by a reinforcement learning framework where the Critic offers natural language critiques instead of scalar rewards, enabling more nuanced feedback to boost the Reasoner’s capability on complex reasoning tasks. The Critic model is trained using Direct Preference Optimization (DPO), leveraging a preference dataset of critiques ranked by Rule-based Reward (RBR) to enhance its critic capabilities. Evaluation results show that the Critic-V framework significantly outperforms existing methods, including GPT-4V, on 5 out of 8 benchmarks, especially regarding reasoning accuracy and efficiency. Combining a dynamic text-based policy for the Reasoner and constructive feedback from the preference-optimized Critic enables a more reliable and context-sensitive multimodal reasoning process. Our approach provides a promising solution to enhance the reliability of VLMs, improving their performance in real-world reasoning-heavy multimodal applications such as autonomous driving and embodied intelligence. Our data and code are released at https://github.com/kyrieLei/Critic-V.

Creating high-fidelity 3D meshes with arbitrary topology, including open surfaces and complex interiors, remains a significant challenge. Existing implicit field methods often require costly and detail-degrading watertight conversion, while other approaches struggle with high resolutions. This paper introduces SparseFlex, a novel sparse-structured isosurface representation that enables differentiable mesh reconstruction at resolutions up to $1024^3$ directly from rendering losses. SparseFlex combines the accuracy of Flexicubes with a sparse voxel structure, focusing computation on surface-adjacent regions and efficiently handling open surfaces. Crucially, we introduce a frustum-aware sectional voxel training strategy that activates only relevant voxels during rendering, dramatically reducing memory consumption and enabling high-resolution training. This also allows, for the first time, the reconstruction of mesh interiors using only rendering supervision. Building upon this, we demonstrate a complete shape modeling pipeline by training a variational autoencoder (VAE) and a rectified flow transformer for high-quality 3D shape generation. Our experiments show state-of-the-art reconstruction accuracy, with a ~82% reduction in Chamfer Distance and a ~88% increase in F-score compared to previous methods, and demonstrate the generation of high-resolution, detailed 3D shapes with arbitrary topology. By enabling high-resolution, differentiable mesh reconstruction and generation with rendering losses, SparseFlex significantly advances the state-of-the-art in 3D shape representation and modeling.

While accurate and user-friendly Computer-Aided Design (CAD) is crucial for industrial design and manufacturing, existing methods still struggle to achieve this due to their over-simplified representations or architectures incapable of supporting multimodal design requirements. In this paper, we attempt to tackle this problem from both methods and datasets aspects. First, we propose a cascade MAR with topology predictor (CMT), the first multimodal framework for CAD generation based on Boundary Representation (B-Rep). Specifically, the cascade MAR can effectively capture the ``edge-counters-surface'' priors that are essential in B-Reps, while the topology predictor directly estimates topology in B-Reps from the compact tokens in MAR. Second, to facilitate large-scale training, we develop a large-scale multimodal CAD dataset, mmABC, which includes over 1.3 million B-Rep models with multimodal annotations, including point clouds, text descriptions, and multi-view images. Extensive experiments show the superior of CMT in both conditional and unconditional CAD generation tasks. For example, we improve Coverage and Valid ratio by +10.68% and +10.3%, respectively, compared to state-of-the-art methods on ABC in unconditional generation. CMT also improves +4.01 Chamfer on image conditioned CAD generation on mmABC.

We present TAR3D, a novel framework that consists of a 3D-aware Vector Quantized-Variational AutoEncoder (VQ-VAE) and a Generative Pre-trained Transformer (GPT) to generate high-quality 3D assets. The core insight of this work is to migrate the multimodal unification and promising learning capabilities of the next-token prediction paradigm to conditional 3D object generation. To achieve this, the 3D VQ-VAE first encodes a wide range of 3D shapes into a compact triplane latent space and utilizes a set of discrete representations from a trainable codebook to reconstruct fine-grained geometries under the supervision of query point occupancy. Then, the 3D GPT, equipped with a custom triplane position embedding called TriPE, predicts the codebook index sequence with prefilling prompt tokens in an autoregressive manner so that the composition of 3D geometries can be modeled part by part. Extensive experiments on ShapeNet and Objaverse demonstrate that TAR3D can achieve superior generation quality over existing methods in text-to-3D and image-to-3D tasks

Scientific discovery contributes largely to the prosperity of human society, and recent progress shows that LLMs could potentially catalyst the process. However, it is still unclear whether LLMs can discover novel and valid hypotheses in chemistry. In this work, we investigate this main research question: whether LLMs can automatically discover novel and valid chemistry research hypotheses, given only a research question? With extensive discussions with chemistry experts, we adopt the assumption that a majority of chemistry hypotheses can be resulted from a research background question and several inspirations. With this key insight, we break the main question into three smaller fundamental questions. In brief, they are: (1) given a background question, whether LLMs can retrieve good inspirations; (2) with background and inspirations, whether LLMs can lead to hypothesis; and (3) whether LLMs can identify good hypotheses to rank them higher. To investigate these questions, we construct a benchmark consisting of 51 chemistry papers published in Nature or a similar level in 2024 (all papers are only available online since 2024). Every paper is divided by chemistry PhD students into three components: background, inspirations, and hypothesis. The goal is to rediscover the hypothesis given only the background and a large chemistry literature corpus consisting the ground truth inspiration papers, with LLMs trained with data up to 2023. We also develop an LLM-based multi-agent framework that leverages the assumption, consisting of three stages reflecting the more smaller questions. The proposed method can rediscover many hypotheses with very high similarity with the ground truth ones, covering the main innovations.

Spatial intelligence is the ability of a machine to perceive, reason, and act in three dimensions within space and time. Recent advancements in large-scale auto-regressive models have demonstrated remarkable capabilities across various reasoning tasks. However, these models often struggle with fundamental aspects of spatial reasoning, particularly in answering questions like "Where am I?" and "What will I see?". While some attempts have been done, existing approaches typically treat them as separate tasks, failing to capture their interconnected nature. In this paper, we present **G**enerative **S**patial **T**ransformer (GST), a novel auto-regressive framework that jointly addresses spatial localization and view prediction. Our model simultaneously estimates the camera pose from a single image and predicts the view from a new camera pose, effectively bridging the gap between spatial awareness and visual prediction. The proposed innovative camera tokenization method enables the model to learn the joint distribution of 2D projections and their corresponding spatial perspectives in an auto-regressive manner. This unified training paradigm demonstrates that joint optimization of pose estimation and novel view synthesis leads to improved performance in both tasks, for the first time, highlighting the inherent relationship between spatial awareness and visual prediction.

Precipitation nowcasting plays a pivotal role in socioeconomic sectors, especially in severe convective weather warnings. Although notable progress has been achieved by approaches mining the spatiotemporal correlations with deep learning, these methods still suffer severe blurriness as the lead time increases, which hampers accurate predictions for extreme precipitation. To alleviate blurriness, researchers explore generative methods conditioned on blurry predictions. However, the pairs of blurry predictions and corresponding ground truth need to be given in advance, making the training pipeline cumbersome and limiting the generality of generative models within blurry modes that appear in training data. By rethinking the blurriness in precipitation nowcasting as a blur kernel acting on predictions, we propose an unsupervised postprocessing method to eliminate the blurriness without the requirement of training with the pairs of blurry predictions and corresponding ground truth. Specifically, we utilize blurry predictions to guide the generation process of a pre-trained unconditional denoising diffusion probabilistic model (DDPM) to obtain high-fidelity predictions with eliminated blurriness. A zero-shot blur kernel estimation mechanism and an auto-scale denoise guidance strategy are introduced to adapt the unconditional DDPM to any blurriness modes varying from datasets and lead times in precipitation nowcasting. Extensive experiments are conducted on 7 precipitation radar datasets, demonstrating the generality and superiority of our method.

Reconstructing 3D scenes from multiple viewpoints is a fundamental task in stereo vision. Recently, advances in generalizable 3D Gaussian Splatting have enabled high-quality novel view synthesis for unseen scenes from sparse input views by feed-forward predicting per-pixel Gaussian parameters without extra optimization. However, existing methods typically generate single-scale 3D Gaussians, which lack representation of both large-scale structure and texture details, resulting in mislocation and artefacts. In this paper, we propose a novel framework, HiSplat, which introduces a hierarchical manner in generalizable 3D Gaussian Splatting to construct hierarchical 3D Gaussians via a coarse-to-fine strategy. Specifically, HiSplat generates large coarse-grained Gaussians to capture large-scale structures, followed by fine-grained Gaussians to enhance delicate texture details. To promote inter-scale interactions, we propose an Error Aware Module for Gaussian compensation and a Modulating Fusion Module for Gaussian repair. Our method achieves joint optimization of hierarchical representations, allowing for novel view synthesis using only two-view reference images. Comprehensive experiments on various datasets demonstrate that HiSplat significantly enhances reconstruction quality and cross-dataset generalization compared to prior single-scale methods. The corresponding ablation study and analysis of different-scale 3D Gaussians reveal the mechanism behind the effectiveness. Code is at https://github.com/Open3DVLab/HiSplat.

Large-scale datasets have been pivotal to the advancements of deep learning models in recent years, but training on such large datasets inevitably incurs substantial storage and computational overhead. Meanwhile, real-world datasets often contain redundant and noisy data, imposing a negative impact on training efficiency and model performance. Data selection has shown promise in identifying the most representative samples from the entire dataset, which aims to minimize the performance gap with reduced training costs. Existing works typically rely on single-modality information to assign importance scores for individual samples, which may lead to inaccurate assessments, especially when dealing with noisy or corrupted samples. To address this limitation, we propose a novel CLIP-powered data selection framework that leverages multimodal information for more robust and generalizable sample selection. Specifically, our framework consists of three key modules—dataset adaptation, sample scoring, and selection optimization—that together harness extensive pre-trained multimodal knowledge to comprehensively assess sample influence and optimize the selection results through multi-objective optimization. Extensive experiments demonstrate that our approach consistently outperforms existing state-of-the-art baselines on various benchmark datasets. Notably, our method effectively removes noisy or damaged samples from the dataset, enabling it to achieve even higher performance with less data. This indicates that it is not only a way to accelerate training but can also improve overall data quality. The implementation is available at https://github.com/Jackbrocp/clip-powered-data-selection.

Video depth estimation has long been hindered by the scarcity of consistent and scalable ground truth data, leading to inconsistent and unreliable results. In this paper, we introduce Depth Any Video, a model that tackles the challenge through two key innovations. First, we develop a scalable synthetic data pipeline, capturing real-time video depth data from diverse virtual environments, yielding 40,000 video clips of 5-second duration, each with precise depth annotations. Second, we leverage the powerful priors of generative video diffusion models to handle real-world videos effectively, integrating advanced techniques such as rotary position encoding and flow matching to further enhance flexibility and efficiency. Unlike previous models, which are limited to fixed-length video sequences, our approach introduces a novel mixed-duration training strategy that handles videos of varying lengths and performs robustly across different frame rates—even on single frames. At inference, we propose a depth interpolation method that enables our model to infer high-resolution video depth across sequences of up to 150 frames. Our model outperforms all previous generative depth models in terms of spatial accuracy and temporal consistency. The code and model weights are open-sourced.

The Earth's weather system involves intricate weather data modalities and diverse weather understanding tasks, which hold significant value to human life. Existing data-driven models focus on single weather understanding tasks (e.g., weather forecasting). While these models have achieved promising results, they fail to tackle various complex tasks within a single and unified model. Moreover, the paradigm that relies on limited real observations for a single scenario hinders the model's performance upper bound. Inspired by the in-context learning paradigm from visual foundation models and large language models, in this paper, we introduce the first generalist weather generalist foundation model (WeatherGFM) to address weather understanding tasks in a unified manner. Specifically, we first unify the representation and definition for diverse weather understanding tasks. Subsequently, we design weather prompt formats to handle different weather data modalities, including single, multiple, and temporal modalities. Finally, we adopt a visual prompting question-answering paradigm for the training of unified weather understanding tasks. Extensive experiments indicate that our WeatherGFM can effectively handle up to 12 weather understanding tasks, including weather forecasting, super-resolution, weather image translation, and post-processing. Our method also showcases generalization ability on unseen tasks. The source code is available at https://github.com/xiangyu-mm/WeatherGFM.

Brain decoding aims to reconstruct visual perception of human subject from fMRI signals, which is crucial for understanding brain’s perception mechanisms. Existing methods are confined to the single-subject paradigm due to substantial brain variability, which leads to weak generalization across individuals and incurs high training costs, exacerbated by limited availability of fMRI data. To address these challenges, we propose MindAligner, an explicit functional alignment framework for cross-subject brain decoding from limited fMRI data. The proposed MindAligner enjoys several merits. First, we learn a Brain Transfer Matrix (BTM) that projects the brain signals of an arbitrary new subject to one of the known subjects, enabling seamless use of pre-trained decoding models. Second, to facilitate reliable BTM learning, a Brain Functional Alignment module is proposed to perform soft cross-subject brain alignment under different visual stimuli with a multi-level brain alignment loss, uncovering fine-grained functional correspondences with high interpretability. Experiments indicate that MindAligner not only outperforms existing methods in visual decoding under data-limited conditions, but also provides valuable neuroscience insights in cross-subject functional analysis. The code will be made publicly available.

Individual brains exhibit striking structural and physiological heterogeneity, yet neural circuits can generate remarkably consistent functional properties across individuals, an apparent paradox in neuroscience. While recent studies have observed preserved neural representations in motor cortex through manual alignment across subjects, the zero-shot validation of such preservation and its generalization to more cortices remain unexplored. Here we present PNBA (Probabilistic Neural-Behavioral Representation Alignment), a new framework that leverages probabilistic modeling to address hierarchical variability across trials, sessions, and subjects, with generative constraints preventing representation degeneration. By establishing reliable cross-modal representational alignment, PNBA reveals robust preserved neural representations in monkey primary motor cortex (M1) and dorsal premotor cortex (PMd) through zero-shot validation. We further establish similar representational preservation in mouse primary visual cortex (V1), reflecting a general neural basis. These findings resolve the paradox of neural heterogeneity by establishing zero-shot preserved neural representations across cortices and species, enriching neural coding insights and enabling zero-shot behavior decoding.

Human-aware navigation (HAN) aims to build autonomous agents that robustly and naturally navigate in human-centered environments. Due to the complex and dynamic nature of this task, existing approaches typically rely on sophisticated pipelines that separately process perception and decision-making to solve it. In this work, we propose an Obstruction Distance Vector based End-to-End Model (ODVEEM), using monocular vision for navigation around humans. The Obstruction Distance Vector (ODV) is an intermediate representation in our model, leveraged to describe the Obstruction Distance to the first future collision in all possible directions in the horizontal field of view. As ODV cannot be calculated directly in the real world, we design a neural network for ODV estimation, formulating it as a classification problem with auxiliary proxy tasks, which play a key role in effectively predicting the implicit future motion of nearby humans. Taking advantage of ODV, ODVEEM supervised by human behavioral heuristics is employed to guide the agent to reach a goal efficiently and avoid potential collisions. Several challenging experiments show our method's substantial improvement over a number of baseline methods, attaining solid performance with zero-shot transfer to unseen simulated and real-world environments.

Human understanding and generation are critical for modeling digital humans and humanoid embodiments. Recently, Human-centric Foundation Models (HcFMs)—inspired by the success of generalist models such as large language and vision models—have emerged to unify diverse human-centric tasks into a single framework, surpassing traditional task-specific approaches. In this survey, we present a comprehensive overview of HcFMs by proposing a taxonomy that categorizes current approaches into four groups: (1) Human-centric Perception Foundation Models that capture fine-grained features for multi-modal 2D and 3D understanding; (2) Human-centric AIGC Foundation Models that generate high-fidelity, diverse human-related content; (3) Unified Perception and Generation Models that integrate these capabilities to enhance both human understanding and synthesis; and (4) Human-centric Agentic Foundation Models that extend beyond perception and generation to learn human-like intelligence and interactive behaviors for humanoid embodied tasks. We review state-of-the-art techniques, discuss emerging challenges and future research directions. This survey aims to serve as a roadmap for researchers and practitioners working towards more robust, versatile, and intelligent digital human and embodiments modeling. Website is https://github.com/HumanCentricModels/Awesome-Human-Centric-Foundation-Models/

Achieving generalizability in solving out-of-distribution tasks is one of the ultimate goals of learning robotic manipulation. Recent progress of Vision-Language Models (VLMs) has shown that VLM-based task planners can alleviate the difficulty of solving novel tasks, by decomposing the compounded tasks as a plan of sequentially executing primitive-level skills that have been already mastered. It is also promising for robotic manipulation to adapt such composable generalization ability, in the form of composable generalization agents (CGAs). However, the community lacks of reliable design of primitive skills and a sufficient amount of primitive-level data annotations. Therefore, we propose RH20T-P, a primitive-level robotic manipulation dataset, which contains about 38k video clips covering 67 diverse manipulation tasks in real-world scenarios. Each clip is manually annotated according to a set of meticulously designed primitive skills that are common in robotic manipulation. Furthermore, we standardize a plan-execute CGA paradigm and implement an exemplar baseline called RA-P on our RH20T-P, whose positive performance on solving unseen tasks validates that the proposed dataset can offer composable generalization ability to robotic manipulation agents. Project homepage: https://sites.google.com/view/rh20t-primitive/main.

Di Zhang, Jianbo Wu, Jingdi Lei, Tong Che, Jiatong Li, Tong Xie, Xiaoshui Huang, Shufei Zhang, Marco Pavone, Yuqiang Li, Wanli Ouyang, Dongzhan Zhou. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Competitive programming is widely used to evaluate the coding and reasoning abilities of large language models. However, the growing presence of duplicate or highly similar problems raises concerns not only about competition fairness, but also about the validity of competitive programming as a benchmark for model evaluation. We introduce a retrieval-oriented benchmark suite for competitive programming, covering four retrieval tasks—two code-centric (Text-to-Code, Code-to-Code) and two newly proposed problem-centric tasks (Problem-to-Duplicate, Simplified-to-Full)—built from a combination of automatically crawled problem–solution data and manually curated annotations. Our contribution includes both high-quality training data and temporally separated test sets for reliable evaluation. We develop two task-specialized retrievers based on this dataset: CPRetriever-Code, trained with a novel Group-InfoNCE loss for problem–code alignment, and CPRetriever-Prob, fine-tuned for problem-level similarity. Both models achieve strong results and are open-sourced for local use. Finally, we analyze LiveCodeBench and find that high-similarity problems inflate model pass rates and reduce differentiation, underscoring the need for similarity-aware evaluation in future benchmarks.

Scientific embodied agents play a crucial role in modern laboratories by automating complex experimental workflows.Compared to typical household environments, laboratory settings impose significantly higher demands on perception of physical-chemical transformations and long-horizon planning, making them an ideal testbed for advancing embodied intelligence.However, its development has been long hampered by the lack of suitable simulator and benchmarks.In this paper, we address this gap by introducing LabUtopia, a comprehensive simulation and benchmarking suite designed to facilitate the development of generalizable, reasoning-capable embodied agents in laboratory settings. Specifically, it integrates i) LabSim, a high-fidelity simulator supporting multi-physics and chemically meaningful interactions; ii) LabScene, a scalable procedural generator for diverse scientific scenes; and iii) LabBench, a hierarchical benchmark spanning five levels of complexity from atomic actions to long-horizon mobile manipulation. LabUtopia supports 30 distinct tasks and includes more than 200 scene and instrument assets, enabling large-scale training and principled evaluation in high-complexity environments.We demonstrate that LabUtopia offers a powerful platform for advancing the integration of perception, planning, and control in scientific-purpose agents and provides a rigorous testbed for exploring the practical capabilities and generalization limits of embodied intelligence in future research. Project web page: https://rui-li023.github.io/labutopia-site/

Recent Vision-Language Models (VLMs) have shown strong performance in general-purpose visual understanding and reasoning, but their ability to comprehend the visual grammar of movie shots remains underexplored and insufficiently evaluated. To bridge this gap, we present \textbf{ShotBench}, a dedicated benchmark for assessing VLMs’ understanding of cinematic language. ShotBench includes 3,049 still images and 500 video clips drawn from more than 200 films, with each sample annotated by trained annotators or curated from professional cinematography resources, resulting in 3,608 high-quality question-answer pairs. We conduct a comprehensive evaluation of over 20 state-of-the-art VLMs across eight core cinematography dimensions. Our analysis reveals clear limitations in fine-grained perception and cinematic reasoning of current VLMs. To improve VLMs capability in cinematography understanding, we construct a large-scale multimodal dataset, named ShotQA, which contains about 70k Question-Answer pairs derived from movie shots. Besides, we propose ShotVL and train this VLM model with a two-stage training strategy, integrating both supervised fine-tuning and Group Relative Policy Optimization (GRPO). Experimental results demonstrate that our model achieves substantial improvements, surpassing all existing strongest open-source and proprietary models evaluated on ShotBench, establishing a new state-of-the-art performance.

We propose Flow-GRPO, the first method to integrate online policy gradient reinforcement learning (RL) into flow matching models. Our approach uses two key strategies: (1) an ODE-to-SDE conversion that transforms a deterministic Ordinary Differential Equation (ODE) into an equivalent Stochastic Differential Equation (SDE) that matches the original model's marginal distribution at all timesteps, enabling statistical sampling for RL exploration; and (2) a Denoising Reduction strategy that reduces training denoising steps while retaining the original number of inference steps, significantly improving sampling efficiency without sacrificing performance. Empirically, Flow-GRPO is effective across multiple text-to-image tasks. For compositional generation, RL-tuned SD3.5-M generates nearly perfect object counts, spatial relations, and fine-grained attributes, increasing GenEval accuracy from $63\%$ to $95\%$. In visual text rendering, accuracy improves from $59\%$ to $92\%$, greatly enhancing text generation. Flow-GRPO also achieves substantial gains in human preference alignment. Notably, very little reward hacking occurred, meaning rewards did not increase at the cost of appreciable image quality or diversity degradation.

Video generation has achieved significant advances through rectified flow techniques, but issues like unsmooth motion and misalignment between videos and prompts persist. In this work, we develop a systematic pipeline that harnesses human feedback to mitigate these problems and refine the video generation model. Specifically, we begin by constructing a large-scale human preference dataset focused on modern video generation models, incorporating pairwise annotations across multi-dimensions. We then introduce VideoReward, a multi-dimensional video reward model, and examine how annotations and various design choices impact its rewarding efficacy. From a unified reinforcement learning perspective aimed at maximizing reward with KL regularization, we introduce three alignment algorithms for flow-based models. These include two training-time strategies: direct preference optimization for flow (Flow-DPO) and reward weighted regression for flow (Flow-RWR), and an inference-time technique, Flow-NRG, which applies reward guidance directly to noisy videos. Experimental results indicate that VideoReward significantly outperforms existing reward models, and Flow-DPO demonstrates superior performance compared to both Flow-RWR and supervised fine-tuning methods. Additionally, Flow-NRG lets users assign custom weights to multiple objectives during inference, meeting personalized video quality needs.

Deciphering how visual stimuli are transformed into cortical responses is a fundamental challenge in computational neuroscience. This visual-to-neural mapping is inherently a one-to-many relationship, as identical visual inputs reliably evoke variable hemodynamic responses across trials, contexts, and subjects. However, existing deterministic methods struggle to simultaneously model this biological variability while capturing the underlying functional consistency that encodes stimulus information. To address these limitations, we propose SynBrain, a generative framework that simulates the transformation from visual semantics to neural responses in a probabilistic and biologically interpretable manner. SynBrain introduces two key components: (i) BrainVAE models neural representations as continuous probability distributions via probabilistic learning while maintaining functional consistency through visual semantic constraints; (ii) A Semantic-to-Neural Mapper acts as a semantic transmission pathway, projecting visual semantics into the neural response manifold to facilitate high-fidelity fMRI synthesis. Experimental results demonstrate that SynBrain surpasses state-of-the-art methods in subject-specific visual-to-fMRI encoding performance. Furthermore, SynBrain adapts efficiently to new subjects with few-shot data and synthesizes high-quality fMRI signals that are effective in improving data-limited fMRI-to-image decoding performance. Beyond that, SynBrain reveals functional consistency across trials and subjects, with synthesized signals capturing interpretable patterns shaped by biological neural variability. Our code is available at https://github.com/MichaelMaiii/SynBrain.

Advances in single-cell sequencing have enabled high-resolution profiling of diverse molecular modalities, while integrating unpaired multi-omics single-cell data remains challenging. Existing approaches either rely on pair information or prior correspondences, or require computing a global pairwise coupling matrix, limiting their scalability and flexibility. In this paper, we introduce a scalable and flexible generative framework called single-cell Multi-omics Regularized Disentangled Representations (scMRDR) for unpaired multi-omics integration. Specifically, we disentangle each cell’s latent representations into modality-shared and modality-specific components using a well-designed $\beta$-VAE architecture, which are augmented with isometric regularization to preserve intra-omics biological heterogeneity, adversarial objective to encourage cross-modal alignment, and masked reconstruction loss strategy to address the issue of missing features across modalities. Our method achieves excellent performance on benchmark datasets in terms of batch correction, modality alignment, and biological signal preservation. Crucially, it scales effectively to large-scale datasets and supports integration of more than two omics, offering a powerful and flexible solution for large-scale multi-omics data integration and downstream biological discovery.

We introduce native-resolution image synthesis, a novel paradigm in generative modeling capable of synthesizing images at arbitrary resolutions and aspect ratios. This approach overcomes the limitations of standard fixed-resolution, square-image methods by inherently handling variable-length visual tokens—a core challenge for conventional techniques. To this end, we propose the Native-resolution diffusion Transformer (NiT), an architecture that explicitly models varying resolutions and aspect ratios within its denoising process. Unconstrained by fixed formats, NiT learns intrinsic visual distributions from images encompassing a wide range of resolutions and aspect ratios. Notably, a single NiT model simultaneously achieves the state-of-the-art performance on both ImageNet-256x256 and 512x512 benchmarks. Surprisingly, akin to the robust zero-shot capabilities seen in advanced Large Language Models, NiT, pretrained solely on ImageNet, demonstrates excellent zero-shot generalization performance. It successfully generates high-fidelity images at previously unseen high resolutions (e.g., 1024x1024, 1536x1536) and diverse aspect ratios (e.g., 16:9,3:1, 4:3), as shown in Figure 1. These findings indicate the significant potential of native-resolution modeling as a bridge between visual generative modeling and advanced LLM methodologies.

With the rise of the fine-tuned–pretrained paradigm, storing numerous fine-tuned models for multi-tasking creates significant storage overhead. Delta compression alleviates this by storing only the pretrained model and the highly compressed delta weights (the differences between fine-tuned and pretrained model weights). However, existing methods fail to maintain both high compression and performance, and often rely on data. To address these challenges, we propose UltraDelta, the first data-free delta compression pipeline that achieves both ultra-high compression and strong performance. UltraDelta is designed to minimize redundancy, maximize information, and stabilize performance across inter-layer, intra-layer, and global dimensions, using three key components: (1) Variance-Based Mixed Sparsity Allocation assigns sparsity based on variance, giving lower sparsity to high-variance layers to preserve inter-layer information. (2) Distribution-Aware Compression applies uniform quantization and then groups parameters by value, followed by group-wise pruning, to better preserve intra-layer distribution. (3) Trace-Norm-Guided Rescaling uses the trace norm of delta weights to estimate a global rescaling factor, improving model stability under higher compression. Extensive experiments across (a) large language models (fine-tuned on LLaMA-2 7B and 13B) with up to 50$\times$ compression, (b) general NLP models (RoBERTa-base, T5-base) with up to 224$\times$ compression, (c) vision models (ViT-B/32, ViT-L/14) with up to 132$\times$ compression, and (d) multi-modal models (BEiT-3) with 18$\times$ compression, demonstrate that UltraDelta consistently outperforms existing methods, especially under ultra-high compression. Code is available at https://github.com/xiaohuiwang000/UltraDelta.

Retrieval-augmented generation (RAG) has become a widely adopted paradigm for enabling knowledge-grounded large language models (LLMs). However, standard RAG pipelines often fail to ensure that model reasoning remains consistent with the evidence retrieved, leading to factual inconsistencies or unsupported conclusions. In this work, we reinterpret RAG as \textit{Retrieval-Augmented Reasoning} and identify a central but underexplored problem: \textit{Reasoning Misalignment}—the divergence between an LLM's internal reasoning trajectory and the evidential constraints provided by retrieval. To address this issue, we propose \textsc{AlignRAG}, a novel iterative framework grounded in \textit{Critique-Driven Alignment (CDA)}. We further introduce \textsc{AlignRAG-auto}, an autonomous variant that dynamically terminates refinement, removing the need to pre-specify the number of critique iterations. At the heart of \textsc{AlignRAG} lies a \textit{contrastive critique synthesis} mechanism that generates retrieval-sensitive critiques while mitigating self-bias. This mechanism trains a dedicated retrieval-augmented \textit{Critic Language Model (CLM)} using labeled critiques that distinguish between evidence-aligned and misaligned reasoning. Empirical evaluations show that our approach significantly improves reasoning fidelity. Our 8B-parameter CLM improves performance over the Self-Refine baseline by \textbf{12.1\%} on out-of-domain tasks and outperforms a standard 72B-parameter CLM by \textbf{2.2\%}. Furthermore, \textsc{AlignRAG-auto} achieves this state-of-the-art performance while dynamically determining the optimal number of refinement steps, enhancing efficiency and usability. \textsc{AlignRAG} remains compatible with existing RAG architectures as a \textit{plug-and-play} module and demonstrates strong robustness under both informative and noisy retrieval scenarios. Overall, \textsc{AlignRAG} offers a principled solution for aligning model reasoning with retrieved evidence, substantially improving the factual reliability and robustness of RAG systems. Our source code is provided at \href{https://github.com/upup-wei/RAG-ReasonAlignment}{link}.

Super-resolution (SR) advances astronomical imaging by enabling cost-effective high-resolution capture, crucial for detecting faraway celestial objects and precise structural analysis. However, existing datasets for astronomical SR (ASR) exhibit three critical limitations: flux inconsistency, object-crop setting, and insufficient data diversity, significantly impeding ASR development. We propose STAR, a large-scale astronomical SR dataset containing 54,738 flux-consistent star field image pairs covering wide celestial regions. These pairs combine Hubble Space Telescope high-resolution observations with physically faithful low-resolution counterparts generated through a flux-preserving data generation pipeline, enabling systematic development of field-level ASR models. To further empower the ASR community, STAR provides a novel Flux Error (FE) to evaluate SR models in physical view. Leveraging this benchmark, we propose a Flux-Invariant Super Resolution (FISR) model that could accurately infer the flux-consistent high-resolution images from input photometry, suppressing several SR state-of-the-art methods by 24.84% on a novel designed flux consistency metric, showing the priority of our method for astrophysics. Extensive experiments demonstrate the effectiveness of our proposed method and the value of our dataset. Code and models are available at https://github.com/GuoCheng12/STAR.

Large language models (LLMs) have shown promise in automating scientific hypothesis generation, yet existing approaches primarily yield coarse-grained hypotheses lacking critical methodological and experimental details. We introduce and formally define the new task of fine-grained scientific hypothesis discovery, which entails generating detailed, experimentally actionable hypotheses from coarse initial research directions. We frame this as a combinatorial optimization problem and investigate the upper limits of LLMs' capacity to solve it when maximally leveraged. Specifically, we explore four foundational questions: (1) how to best harness an LLM's internal heuristics to formulate the fine-grained hypothesis it itself would judge as the most promising among all the possible hypotheses it might generate, based on its own internal scoring-thus defining a latent reward landscape over the hypothesis space; (2) whether such LLM-judged better hypotheses exhibit stronger alignment with ground-truth hypotheses; (3) whether shaping the reward landscape using an ensemble of diverse LLMs of similar capacity yields better outcomes than defining it with repeated instances of the strongest LLM among them; and (4) whether an ensemble of identical LLMs provides a more reliable reward landscape than a single LLM. To address these questions, we propose a hierarchical search method that incrementally proposes and integrates details into the hypothesis, progressing from general concepts to specific experimental configurations. We show that this hierarchical process smooths the reward landscape and enables more effective optimization. Empirical evaluations on a new benchmark of expert-annotated fine-grained hypotheses from recent literature show that our method consistently outperforms strong baselines.

In-silico prediction of protein mutant stability, measured by the difference in Gibbs free energy change ($\Delta \Delta G$), is fundamental for protein engineering. Current sequence-to-label methods typically employ two-stage pipelines: (i) encoding mutant sequences using neural networks (e.g., transformers), followed by (ii) the $\Delta \Delta G$ regression from the latent representations. Although these methods have demonstrated promising performance, their dependence on specialized neural network encoders significantly increases the complexity. Additionally, the requirement to compute latent representations individually for each mutant sequence negatively impacts computational efficiency and poses the risk of overfitting. This work proposes the Venus-MAXWELL framework, which reformulates mutation $\Delta \Delta G$ prediction as a sequence-to-landscape task. In Venus-MAXWELL, mutations of a protein and their corresponding $\Delta \Delta G$ values are organized into a landscape matrix, allowing our framework to learn the $\Delta \Delta G$ landscape of a protein with a single forward and backward pass during training. To this end, we curated a new $\Delta \Delta G$ benchmark dataset with strict controls on data leakage and redundancy to ensure robust evaluation. Leveraging the zero-shot scoring capability of protein language models (PLMs), Venus-MAXWELL effectively utilizes the evolutionary patterns learned by PLMs during pre-training. More importantly, Venus-MAXWELL is compatible with multiple protein language models. For example, when integrated with the ESM-IF, Venus-MAXWELL achieves higher accuracy than ThermoMPNN with 10$\times$ faster in inference speed (despite having 50$\times$ more parameters than ThermoMPNN). The training codes, model weights, and datasets are publicly available at https://github.com/ai4protein/Venus-MAXWELL.

Recent studies have demonstrated the importance of high-quality visual representations in image generation and have highlighted the limitations of generative models in image understanding. As a generative paradigm originally designed for natural language, autoregressive models face similar challenges. In this work, we present the first systematic investigation into the mechanisms of applying the next-token prediction paradigm to the visual domain. We identify three key properties that hinder the learning of high-level visual semantics: local and conditional dependence, inter-step semantic inconsistency, and spatial invariance deficiency. We show that these issues can be effectively addressed by introducing self-supervised objectives during training, leading to a novel training framework, Self-guided Training for AutoRegressive models (ST-AR). Without relying on pre-trained representation models, ST-AR significantly enhances the image understanding ability of autoregressive models and leads to improved generation quality. Specifically, ST-AR brings approximately 42% FID improvement for LlamaGen-L and 49% FID improvement for LlamaGen-XL, while maintaining the same sampling strategy.

Recent advances in Multimodal Large Language Models (MLLMs) have significantly improved 2D visual understanding, prompting interest in their application to complex 3D reasoning tasks. However, it remains unclear whether these models can effectively capture the detailed spatial information required for robust real-world performance, especially cross-view consistency, a key requirement for accurate 3D reasoning. Considering this issue, we introduce Viewpoint Learning, a task designed to evaluate and improve the spatial reasoning capabilities of MLLMs. We present the Viewpoint-100K dataset, consisting of 100K object-centric image pairs with diverse viewpoints and corresponding question-answer pairs. Our approach employs a two-stage fine-tuning strategy: first, foundational knowledge is injected to the baseline MLLM via Supervised Fine-Tuning (SFT) on Viewpoint-100K, resulting in significant improvements across multiple tasks; second, generalization is enhanced through Reinforcement Learning using the Group Relative Policy Optimization (GRPO) algorithm on a broader set of questions. Additionally, we introduce a hybrid cold-start initialization method designed to simultaneously learn viewpoint representations and maintain coherent reasoning thinking. Experimental results show that our approach significantly activates the spatial reasoning ability of MLLM, improving performance on both in-domain and out-of-domain reasoning tasks. Our findings highlight the value of developing foundational spatial skills in MLLMs, supporting future progress in robotics, autonomous systems, and 3D scene understanding.

Aircraft manufacturing is the jewel in the crown of industry, in which generating high-fidelity airfoil geometries with controllable and editable representations remains a fundamental challenge. Existing deep learning methods, which typically rely on predefined parametric representations (e.g., Bézier curves) or discrete point sets, face an inherent trade-off between expressive power and resolution adaptability. To tackle this challenge, we introduce FuncGenFoil, a novel function-space generative model that directly reconstructs airfoil geometries as function curves. Our method inherits the advantages of arbitrary-resolution sampling and smoothness from parametric functions, as well as the strong expressiveness of discrete point-based representations. Empirical evaluations demonstrate that FuncGenFoil improves upon state-of-the-art methods in airfoil generation, achieving a relative 74.4% reduction in label error and a 23.2% increase in diversity on the AF-200K dataset. Our results highlight the advantages of function-space modeling for aerodynamic shape optimization, offering a powerful and flexible framework for high-fidelity airfoil design.

Deep learning-based computational methods have achieved promising results in predicting protein-protein interactions (PPIs). However, existing benchmarks predominantly focus on isolated pairwise evaluations, overlooking a model's capability to reconstruct biologically meaningful PPI networks, which is crucial for biology research. To address this gap, we introduce PRING, the first comprehensive benchmark that evaluates PRotein-protein INteraction prediction from a Graph-level perspective. PRING curates a high-quality, multi-species PPI network dataset comprising 21,484 proteins and 186,818 interactions, with well-designed strategies to address both data redundancy and leakage. Building on this golden-standard dataset, we establish two complementary evaluation paradigms: (1) topology-oriented tasks, which assess intra and cross-species PPI network construction, and (2) function-oriented tasks, including protein complex pathway prediction, GO module analysis, and essential protein justification. These evaluations not only reflect the model's capability to understand the network topology but also facilitate protein function annotation, biological module detection, and even disease mechanism analysis. Extensive experiments on four representative model categories, consisting of sequence similarity-based, naive sequence-based, protein language model-based, and structure-based approaches, demonstrate that current PPI models have potential limitations in recovering both structural and functional properties of PPI networks, highlighting the gap in supporting real-world biological applications. We believe PRING provides a reliable platform to guide the development of more effective PPI prediction models for the community. The dataset and source code of PRING are available at https://github.com/SophieSarceau/PRING.

Understanding and decoding human brain activity from electroencephalography (EEG) signals is a fundamental problem in neuroscience and artificial intelligence, with applications ranging from cognition and emotion recognition to clinical diagnosis and brain–computer interfaces. While recent EEG foundation models have made progress in generalized brain decoding by leveraging unified architectures and large-scale pretraining, they inherit a scale-agnostic dense modeling paradigm from NLP and vision. This design overlooks an intrinsic property of neural activity—cross-scale spatiotemporal structure. Different EEG task patterns span a broad range of temporal and spatial scales, from brief neural activations to slow-varying rhythms, and from localized cortical activations to large-scale distributed interactions. Ignoring this diversity may lead to suboptimal representations and weakened generalization ability. To address these limitations, we propose CSBrain, a Cross-scale Spatiotemporal Brain foundation model for generalized EEG decoding. CSBrain introduces two key components: (i) Cross-scale Spatiotemporal Tokenization (CST), which aggregates multi-scale features within localized temporal windows and anatomical brain regions into compact scale-aware token representations; and (ii) Structured Sparse Attention (SSA), which models cross-window and cross-region dependencies for diverse decoding tasks, further enriching scale diversities while eliminating the spurious dependencies. CST and SSA are alternately stacked to progressively integrate cross-scale spatiotemporal dependencies. Extensive experiments across 11 representative EEG tasks and 16 datasets demonstrate that CSBrain consistently outperforms both task-specific models and strong foundation baselines. These results establish cross-scale modeling as a key inductive bias for generalized EEG decoding and highlight CSBrain as a robust backbone for future brain–AI research.

Inspired by generative paradigms in image and video, 3D shape generation has made notable progress, enabling the rapid synthesis of high-fidelity 3D assets from a single image. However, current methods still face challenges, including the lack of intricate details, overly smoothed surfaces, and fragmented thin-shell structures. These limitations leave the generated 3D assets still one step short of meeting the standards favored by artists. In this paper, we present ShapeGen, which achieves high-quality image-to-3D shape generation through 3D representation and supervision improvements, resolution scaling up, and the advantages of linear transformers. These advancements allow the generated assets to be seamlessly integrated into 3D pipelines, facilitating their widespread adoption across various applications. Specifically, in contrast to existing methods: 1) We investigate how different representations and VAE supervision strategies affect the generation process, and address issues like aliasing artifacts and fragmented thin-shell structures by using an TSDF-based representation supervised with BCE loss. 2) We scale up the resolution of 3D data, image conditioning inputs, and the number of latent tokens to enhance generation fidelity. 3) We adopt mixed conditioning using raw RGB images and normal maps during training, effectively resolving ambiguities caused by inconsistencies between ControlNet-generated RGB images and the underlying geometry from untextured assets. 4) We replace the original softmax attention with linear attention to improve training and inference efficiency when handling a large number of latent tokens. 5) We introduce an inference-time scaling strategy that enhances generation quality at test time. Through extensive experiments, we validate the impact of these improvements on overall performance. Ultimately, thanks to the synergistic effects of these enhancements, ShapeGen achieves a significant leap in image-to-3D generation, establishing a new state-of-the-art performance.

Effective multi-shot generation demands purposeful, film-like transitions and strict cinematic continuity. Current methods, however, often prioritize basic visual consistency, neglecting crucial editing patterns (e.g., shot/reverse shot, cutaways) that drive narrative flow for compelling storytelling. This yields outputs that may be visually coherent but lack narrative sophistication and true cinematic integrity. To bridge this, we introduce Next Shot Generation (NSG): synthesizing a subsequent, high-quality shot that critically conforms to professional editing patterns while upholding rigorous cinematic continuity. Our framework, Cut2Next, leverages a Diffusion Transformer (DiT). It employs in-context tuning guided by a novel Hierarchical Multi-Prompting strategy. This strategy uses Relational Prompts to define overall context and inter-shot editing styles. Individual Prompts then specify per-shot content and cinematographic attributes. Together, these guide Cut2Next to generate cinematically appropriate next shots. Architectural innovations, Context-Aware Condition Injection (CACI) and Hierarchical Attention Mask (HAM), further integrate these diverse signals without introducing new parameters. We construct RawCuts (large-scale) and CuratedCuts (refined) datasets, both with hierarchical prompts, and introduce CutBench for evaluation. Experiments show Cut2Next excels in visual consistency and text fidelity. Crucially, user studies reveal a strong preference for Cut2Next, particularly for its adherence to intended editing patterns and overall cinematic continuity, validating its ability to generate high-quality, narratively expressive, and cinematically coherent subsequent shots.

Protein sequence not only determines its structure but also provides important clues of its subcellular localization. Although a series of artificial intelligence models have been reported to predict protein subcellular localization, most of them provide only textual outputs. Here, we present deepGPS, a deep generative model for protein subcellular localization prediction. After training with protein primary sequences and fluorescence images, deepGPS shows the ability to predict cytoplasmic and nuclear localizations by reporting both textual labels and generative images as outputs. In addition, cell-type-specific deepGPS models can be developed by using distinct image datasets from different cell lines for comparative analyses. Moreover, deepGPS shows potential to be further extended for other specific organelles, such as vesicles and endoplasmic reticulum, even with limited volumes of training data. Finally, the openGPS website (https://bits.fudan.edu.cn/opengps) is constructed to provide a publicly accessible and user-friendly platform for studying protein subcellular localization and function.

Recently, person re-identification (ReID) has witnessed fast development due to its broad practical applications and proposed various settings, e.g., traditional ReID, clothes-changing ReID, and visible-infrared ReID. However, current studies primarily focus on single specific tasks, which limits model applicability in real-world scenarios. This paper aims to address this issue by introducing a novel instruct-ReID task that unifies 6 existing ReID tasks in one model and retrieves images based on provided visual or textual instructions. Instruct-ReID is the first exploration of a general ReID setting, where 6 existing ReID tasks can be viewed as special cases by assigning different instructions. To facilitate research in this new instruct-ReID task, we propose a large-scale OmniReID++ benchmark equipped with diverse data and comprehensive evaluation methods, e.g., task-specific and task-free evaluation settings. In the task-specific evaluation setting, gallery sets are categorized according to specific ReID tasks. We propose a novel baseline model, IRM, with an adaptive triplet loss to handle various retrieval tasks within a unified framework. For task-free evaluation setting, where target person images are retrieved from task-agnostic gallery sets, we further propose a new method called IRM++ with novel memory bank-assisted learning. Extensive evaluations of IRM and IRM++ on OmniReID++ benchmark demonstrate the superiority of our proposed methods, achieving state-of-the-art performance on 10 test sets.

Geometry plays a significant role in monocular 3D object detection. It can be used to estimate object depth by using the perspective projection between object's physical size and 2D projection in the image plane, which can introduce mathematical priors into deep models. However, this projection process also introduces error amplification, where the error of the estimated height is amplified and reflected into the projected depth. It leads to unreliable depth inferences and also impairs training stability. To tackle this problem, we propose a novel Geometry Uncertainty Propagation Network (GUPNet++) by modeling geometry projection in a probabilistic manner. This ensures depth predictions are well-bounded and associated with a reasonable uncertainty. The significance of introducing such geometric uncertainty is two-fold: (1). It models the uncertainty propagation relationship of the geometry projection during training, improving the stability and efficiency of the end-to-end model learning. (2). It can be derived to a highly reliable confidence to indicate the quality of the 3D detection result, enabling more reliable detection inference. Experiments show that the proposed approach not only obtains (state-of-the-art) SOTA performance in image-based monocular 3D detection but also demonstrates superiority in efficacy with a simplified framework. The code and model will be released at https://github.com/SuperMHP/GUPNet_Plus.

Pedestrian detection currently suffers from two issues in crowded scenes: occlusion and dense boundary prediction, making it still challenging in complex real-world scenarios. In recent years, Convolutional Neural Networks (CNN) and Vision Transformers (ViT) have shown their superiorities in addressing these issues, where ViTs capture global feature dependency to infer occlusion parts and CNNs make accurate dense predictions by local detailed features. Nevertheless, limited by the narrow receptive field, CNNs fail to infer occlusion parts, while ViTs tend to ignore local features that are vital to distinguish different pedestrians in the crowd. Therefore, it is essential to combine the advantages of CNN and ViT for pedestrian detection. However, manually designing a specific CNN and ViT hybrid network requires enormous time and resources for trial and error. To address this issue, we propose the first Neural Architecture Search (NAS) framework specifically designed for pedestrian detection named NAS-PED, which automatically designs an appropriate CNNs and ViTs hybrid backbone for the crowded pedestrian detection task. Specifically, we formulate transformers and convolutions with various kernel sizes in the same format, which provides an unconstrained space for diverse hybrid network search. Furthermore, to search for a suitable backbone, we propose an information bottleneck based NAS objective function, which treats the process of NAS as an information extraction process, preserving relevant information and suppressing redundant information from the dense pedestrians in crowd scenes Extensive experiments on CrowdHuman, CityPersons and EuroCity Persons datasets demonstrate the effectiveness of the proposed method. Our NAS-PED obtains absolute gains of 4.0% MR and 1.9% AP over the state-of-the-art (SOTA) pedestrian detection framework on CrowdHuman datasets. For the CityPersons and EuroCity Persons datasets, the searched backbone achieves stable improvement across all three subsets, outperforming some large language-image pre-trained models. Code will be released after acceptance.

Human-centric perception tasks, e.g., pedestrian detection, skeleton-based action recognition, and pose estimation, have wide industrial applications, such as metaverse and sports analysis. There is a recent surge to develop human-centric foundation models that can benefit a broad range of human-centric perception tasks. While many human-centric foundation models have achieved success, they did not explore 3D and vision-language tasks for human-centric and required task-specific finetuning. These limitations restrict their application to more downstream tasks and situations. To tackle these problems, we present Hulk, the first multimodal human-centric generalist model, capable of addressing 2D vision, 3D vision, skeleton-based, and vision-language tasks without task-specific finetuning. The key to achieving this is condensing various task-specific heads into two general heads, one for discrete representations, e.g., languages, and the other for continuous representations, e.g., location coordinates. The outputs of two heads can be further stacked into four distinct input and output modalities. This uniform representation enables Hulk to treat diverse human-centric tasks as modality translation, integrating knowledge across a wide range of tasks. Comprehensive evaluations of Hulk on 12 benchmarks covering 8 human-centric tasks demonstrate the superiority of our proposed method, achieving state-of-the-art performance in 11 benchmarks.

Understanding videos, especially aligning them with textual data, presents a significant challenge in computer vision. The advent of vision-language models (VLMs) like CLIP has sparked interest in leveraging their capabilities for enhanced video understanding, showing marked advancements in both performance and efficiency. However, current methods often neglect vital user-generated metadata such as video titles. In this paper, we present Cap4Video++, a universal framework that leverages auxiliary captions to enrich video understanding. More recently, we witness the flourishing of large language models (LLMs) like ChatGPT. Cap4Video++ harnesses the synergy of vision-language models (VLMs) and large language models (LLMs) to generate video captions, utilized in three key phases: (i) Input stage employs Semantic Pair Sampling to extract beneficial samples from captions, aiding contrastive learning. (ii) Intermediate stage sees Video-Caption Cross-modal Interaction and Adaptive Caption Selection work together to bolster video and caption representations. (iii) Output stage introduces a Complementary Caption-Text Matching branch, enhancing the primary video branch by improving similarity calculations. Our comprehensive experiments on text-video retrieval and video action recognition across nine benchmarks clearly demonstrate Cap4Video++'s superiority over existing models, highlighting its effectiveness in utilizing automatically generated captions to advance video understanding.

Residual networks have shown great success and become indispensable in recent deep neural network models. In this work, we aim to re-investigate the training process of residual networks from a novel perspective of loafing, and further propose a new training scheme as well as three improved strategies for boosting residual networks beyond their performance limits. Previous research has suggested that residual networks can be considered as ensembles of shallow networks, which implies that the final performance of a residual network is influenced by a group of subnetworks. Furthermore, we identify a previously overlooked problem, where subnetworks within a residual network are prone to exert less effort when working as part of a group compared to working alone. We define this problem as network loafing. Since network loafing may inevitably cause the sub-par performance of the residual network, we propose a novel training scheme called stimulative training, which randomly samples a residual subnetwork and calculates the KL divergence loss between the sampled subnetwork and the given residual network for extra supervision. In order to unleash the potential of stimulative training, we further propose three simple-yet-effective strategies, including a novel KL- loss that only aligns the network logits direction, random smaller inputs for subnetworks, and inter-stage sampling rules. Comprehensive experiments and analysis verify the effectiveness of stimulative training as well as its three improved strategies. For example, the proposed method can boost the performance of ResNet50 on ImageNet to 80.5% Top1 accuracy without using any extra data, model, trick, or changing the structure. With only uniform augment, the performance can be further improved to 81.0% Top1 accuracy, better than the best training recipes provided by Timm library and PyTorch official version. We also verify its superiority on various typical models, datasets, and tasks and give some theoretical analysis. As such, we advocate utilizing the proposed method as a general and next-generation technology to train residual networks.

In contrast to numerous NLP and 2D vision foundational models, training a 3D foundational model poses considerably greater challenges. This is primarily due to the inherent data variability and diversity of downstream tasks. In this paper, we introduce a novel universal 3D pre-training framework designed to facilitate the acquisition of efficient 3D representations. Considering that informative 3D features should encode rich geometry and appearance cues that can be utilized to render realistic images, we propose to learn 3D representations by differentiable neural rendering. We train a 3D backbone with a volumetric neural renderer by comparing the rendered with the real images. Notably, our pre-trained encoder can be seamlessly applied to various downstream tasks. These tasks include semantic challenges like 3D detection and segmentation, which involve scene understanding, and non-semantic tasks like 3D reconstruction and image synthesis, which focus on geometry and visuals. They span both indoor and outdoor scenarios. We also illustrate the capability of pre-training a 2D backbone using the proposed methodology, surpassing conventional pre-training methods by a large margin. For the first time, PonderV2 achieves state-of-the-art performance on 11 indoor and outdoor benchmarks, implying its effectiveness.

As Vision Transformers (ViTs) become increasingly popular in various vision tasks, one may question:if a new training scheme for ViTs exists that can improve performance without increasing training and inference computation cost?In this paper, we affirmatively answer this question with a novel Sparse-to-Dense (S2D) training scheme. Specifically, we decouple the training and inference phases of ViTs. During training, we replace some Feed-Forward Network (FFN) layers of ViTs with computationally efficient RUP-Mixture-of-FFN (RUP-MoF) layers, each comprising multiple FFN experts and allocating tokens to experts via Random Uniform Partition (RUP). Furthermore, an additional Experts Weights Averaging (EWA) update is performed specifically on these RUP-MoF layers after each gradient update. After training, we convert each RUP-MoF layer back to a single FFN layer by averaging the experts, transforming the training-time sparse model back to the original dense ViT model for inference. We further provide theoretical analysis to illustrate why and how it works. Comprehensive experiments across various 2D and 3D vision tasks, ViT architectures and datasets validate the effectiveness and generalization ability of the proposed S2D training scheme. Besides, we show that, S2D training scheme can also be applied to improve the performance of Transformer-based language models, and EWA update technique can also significantly improve the effectiveness of classic Mixture-of-Experts on various 2D vision small-scale datasets and 3D vision tasks.

Improving the efficiency of Neural Architecture Search (NAS) is a challenging but significant task that has received much attention. Previous studies mainly adopt the Differentiable Architecture Search (DARTS) and improve its search strategies or modules to enhance search efficiency. Recently, some methods have started considering data reduction for speedup, but they are not tightly coupled with the architecture search process and cannot capture the training dynamics of DARTS well, resulting in sub-optimal performances. To this end, this work pioneers an exploration into the critical role of dataset characteristics in the bi-level optimization of DARTS, and then proposes a novel Bi-level Data Pruning (BDP) paradigm that targets the weights and architecture levels of DARTS to enhance efficiency from a data perspective. Specifically, we introduce a progressive bi-level data pruning strategy that utilizes supernet prediction dynamics as the metric to gradually prune unsuitable samples for DARTS during the search. An effective automatic class balance constraint is also integrated into BDP, to suppress potential class imbalances resulting from data-efficient algorithms. Comprehensive evaluations on the NAS-Bench-201 search space, DARTS search space, and MobileNet-like search space validate that BDP reduces search costs by over 50% while achieving superior performance when applied to the baseline DARTS. Besides, we demonstrate that BDP can harmoniously integrate with advanced DARTS variants, like P-DARTS, PC-DARTS, EG-NAS, and$\beta $-DARTS, offering an approximately$2\times $speedup with minimal performance compromise.

The pretrain-finetune paradigm brings about the release of numerous model weights. Under this background, model merging is becoming increasingly popular, as it enables a model to handle multiple tasks by fusing model weights from these tasks, without the need for labeled data, additional training, or high training costs. Though with great potential, model merging suffers from severe performance degradation due to the interference among model weights. And existing model merging methods (i.e., static merging) commonly provide a single set of merging coefficients for all the input samples and do not distinguish layers based on the severity of weight interference, which may not be the optimal solution. In this paper, we propose MoW-Merging, a dynamic model merging method based on Mixture of Weights. First, we apply a gating network to adaptively generate merging coefficients depending on the input samples, realizing sample-wisely dynamic merging and automated classifier selection. The gating network is lightweight and is trained with only a small number of unlabeled data. Further, we utilize a weight similarity metric to judge the severity of weight interference of each layer and apply suitable merging methods to different layers. The proposed MoW-Merging shows plug-and-play capabilities and can be seamlessly combined with various model merging methods to greatly boost their performance. The effectiveness of MoW-Merging is validated by comprehensive experiments on various classical and newly-established benchmarks under multiple settings. The code is available athttps://github.com/harveyhuang18/Mixture_of_Weights.

Most existing weakly supervised semantic segmentation (WSSS) methods rely on class activation mapping (CAM) to extract coarse class-specific localization maps using image-level labels. Prior works have commonly used an off-line heuristic thresholding process that combines the CAM maps with off-the-shelf saliency maps produced by a general pretrained saliency model to produce more accurate pseudo-segmentation labels. We propose AuxSegNet+, a weakly supervised auxiliary learning framework to explore the rich information from these saliency maps and the significant intertask correlation between saliency detection and semantic segmentation. In the proposed AuxSegNet+, saliency detection and multilabel image classification are used as auxiliary tasks to improve the primary task of semantic segmentation with only image-level ground-truth labels. We also propose a cross-task affinity learning mechanism to learn pixel-level affinities from the saliency and segmentation feature maps. In particular, we propose a cross-task dual-affinity learning module to learn both pairwise and unary affinities, which are used to enhance the task-specific features and predictions by aggregating both query-dependent and query-independent global context for both saliency detection and semantic segmentation. The learned cross-task pairwise affinity can also be used to refine and propagate CAM maps to provide better pseudo labels for both tasks. Iterative improvement of segmentation performance is enabled by cross-task affinity learning and pseudo-label updating. Extensive experiments demonstrate the effectiveness of the proposed approach with new state-of-the-art WSSS results on the challenging PASCAL VOC and MS COCO benchmarks.

The pretrain-finetune paradigm has achieved great success in NLP and 2D image fields because of the high-quality representation ability and transferability of their pretrained models. However, pretraining such a strong model is difficult in the 3D point cloud field due to the limited amount of point cloud sequences. This paper introduces Efficient Point Cloud Learning (EPCL), an effective and efficient point cloud learner for directly training high-quality point cloud models with a frozen CLIP transformer. Our EPCL connects the 2D and 3D modalities by semantically aligning the image features and point cloud features without paired 2D-3D data. Specifically, the input point cloud is divided into a series of local patches, which are converted to token embeddings by the designed point cloud tokenizer. These token embeddings are concatenated with a task token and fed into the frozen CLIP transformer to learn point cloud representation. The intuition is that the proposed point cloud tokenizer projects the input point cloud into a unified token space that is similar to the 2D images. Comprehensive experiments on 3D detection, semantic segmentation, classification and few-shot learning demonstrate that the CLIP transformer can serve as an efficient point cloud encoder and our method achieves promising performance on both indoor and outdoor benchmarks. In particular, performance gains brought by our EPCL are 19.7 AP50 on ScanNet V2 detection, 4.4 mIoU on S3DIS segmentation and 1.2 mIoU on SemanticKITTI segmentation compared to contemporary pretrained models. Code is available at \url{https://github.com/XiaoshuiHuang/EPCL}.

De novo peptide sequencing from mass spectrometry (MS) data is a critical task in proteomics research. Traditional de novo algorithms have encountered a bottleneck in accuracy due to the inherent complexity of proteomics data. While deep learning-based methods have shown progress, they reduce the problem to a translation task, potentially overlooking critical nuances between spectra and peptides. In our research, we present ContraNovo, a pioneering algorithm that leverages contrastive learning to extract the relationship between spectra and peptides and incorporates the mass information into peptide decoding, aiming to address these intricacies more efficiently. Through rigorous evaluations on two benchmark datasets, ContraNovo consistently outshines contemporary state-of-the-art solutions, underscoring its promising potential in enhancing de novo peptide sequencing.

Recent research understands the residual networks from a new perspective of the implicit ensemble model. From this view, previous methods such as stochastic depth and stimulative training have further improved the performance of the residual network by sampling and training of its subnets. However, they both use the same supervision for all subnets of different capacities and neglect the valuable knowledge generated by subnets during training. In this manuscript, we mitigate the significant knowledge distillation gap caused by using the same kind of supervision and advocate leveraging the subnets to provide diverse knowledge. Based on this motivation, we propose a group knowledge based training framework for boosting the performance of residual networks. Specifically, we implicitly divide all subnets into hierarchical groups by subnet-in-subnet sampling, aggregate the knowledge of different subnets in each group during training, and exploit upper-level group knowledge to supervise lower-level subnet group. Meanwhile, we also develop a subnet sampling strategy that naturally samples larger subnets, which are found to be more helpful than smaller subnets in boosting performance for hierarchical groups. Compared with typical subnet training and other methods, our method achieves the best efficiency and performance trade-offs on multiple datasets and network structures. The code is at https://github.com/tsj-001/AAAI24-GKT.

Semi-supervised learning aims to leverage numerous unlabeled data to improve the model performance. Current semi-supervised 3D object detection methods typically use a teacher to generate pseudo labels for a student, and the quality of the pseudo labels is essential for the final performance. In this paper, we propose PatchTeacher, which focuses on partial scene 3D object detection to provide high-quality pseudo labels for the student. Specifically, we divide a complete scene into a series of patches and feed them to our PatchTeacher sequentially. PatchTeacher leverages the low memory consumption advantage of partial scene detection to process point clouds with a high-resolution voxelization, which can minimize the information loss of quantization and extract more fine-grained features. However, it is non-trivial to train a detector on fractions of the scene. Therefore, we introduce three key techniques, i.e., Patch Normalizer, Quadrant Align, and Fovea Selection, to improve the performance of PatchTeacher. Moreover, we devise PillarMix, a strong data augmentation strategy that mixes truncated pillars from different LiDAR scans to generate diverse training samples and thus help the model learn more general representation. Extensive experiments conducted on Waymo and ONCE datasets verify the effectiveness and superiority of our method and we achieve new state-of-the-art results, surpassing existing methods by a large margin. Codes are available at https://github.com/LittlePey/PTPM.

Generating realistic human motion from given action descriptions has experienced significant advancements because of the emerging requirement of digital humans. While recent works have achieved impressive results in generating motion directly from textual action descriptions, they often support only a single modality of the control signal, which limits their application in the real digital human industry. This paper presents a Motion General-Purpose generaTor (MotionGPT) that can use multimodal control signals, e.g., text and single-frame poses, for generating consecutive human motions by treating multimodal signals as special input tokens in large language models (LLMs). Specifically, we first quantize multimodal control signals into discrete codes and then formulate them in a unified prompt instruction to ask the LLMs to generate the motion answer. Our MotionGPT demonstrates a unified human motion generation model with multimodal control signals by tuning a mere 0.4% of LLM parameters. To the best of our knowledge, MotionGPT is the first method to generate human motion by multimodal control signals, which we hope can shed light on this new direction. Visit our webpage at https://qiqiapink.github.io/MotionGPT/.

Offline-to-online Reinforcement Learning (O2O RL) aims to improve the performance of offline pretrained policy using only a few online samples. Built on offline RL algorithms, most O2O methods focus on the balance between RL objective and pessimism, or the utilization of offline and online samples. In this paper, from a novel perspective, we systematically study the challenges that remain in O2O RL and identify that the reason behind the slow improvement of the performance and the instability of online finetuning lies in the inaccurate Q-value estimation inherited from offline pretraining. Specifically, we demonstrate that the estimation bias and the inaccurate rank of Q-value cause a misleading signal for the policy update, making the standard offline RL algorithms, such as CQL and TD3-BC, ineffective in the online finetuning. Based on this observation, we address the problem of Q-value estimation by two techniques: (1) perturbed value update and (2) increased frequency of Q-value updates. The first technique smooths out biased Q-value estimation with sharp peaks, preventing early-stage policy exploitation of sub-optimal actions. The second one alleviates the estimation bias inherited from offline pretraining by accelerating learning. Extensive experiments on the MuJoco and Adroit environments demonstrate that the proposed method, named SO2, significantly alleviates Q-value estimation issues, and consistently improves the performance against the state-of-the-art methods by up to 83.1%.

Ge Bai, Jie Liu, Xingyuan Bu, Yancheng He, Jiaheng Liu, Zhanhui Zhou, Zhuoran Lin, Wenbo Su, Tiezheng Ge, Bo Zheng, Wanli Ouyang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Zhanhui Zhou, Jie Liu, Zhichen Dong, Jiaheng Liu, Chao Yang, Wanli Ouyang, Yu Qiao. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

This paper is not motivated to seek innovation within the attention mechanism. Instead, it focuses on overcoming the existing trade-offs between accuracy and efficiency within the context of point cloud processing, leveraging the power of scale. Drawing inspiration from recent advances in 3D large-scale representation learning, we recognize that model performance is more influenced by scale than by intricate design. Therefore, we present Point Transformer V3 (PTv3), which prioritizes simplicity and efficiency over the accuracy of certain mechanisms that are minor to the over-all performance after scaling, such as replacing the precise neighbor search by KNN with an efficient serialized neighbor mapping of point clouds organized with specific patterns. This principle enables significant scaling, expanding the receptive field from 16 to 1024 points while remaining efficient (a 3 x increase in processing speed and a 10 x improvement in memory efficiency compared with its pre-decessor, PTv2). PTv3 attains state-of-the-art results on over 20 downstream tasks that span both indoor and out-door scenarios. Further enhanced with multi-dataset joint training, PTv3 pushes these results to a higher level.

Human intelligence can retrieve any person according to both visual and language descriptions. However, the current computer vision community studies specific person re-identification (ReID) tasks in different scenarios separately, which limits the applications in the real world. This paper strives to resolve this problem by proposing a new instruct-ReID task that requires the model to retrieve images according to the given image or language instructions. Our instruct-ReID is a more general ReID setting, where existing 6 ReID tasks can be viewed as special cases by designing different instructions. We propose a large-scale OmniReID benchmark and an adaptive triplet loss as a baseline method to facilitate research in this new setting. Experimental results show that the proposed multi-purpose ReID model, trained on our OmniReID benchmark without finetuning, can improve +0.5%, +0.6%, +7.7% mAP on Market1501, MSMT17, CUHK03 for traditional ReID, +6.4%, +7.1%, +11.2% mAP on PRCC, VC-Clothes, LTCC for clothes-changing ReID, +11.7% mAP on COCAS+ real2 for clothes template based clothes-changing ReID when using only RGB images, +24.9% mAP on COCAS+ real2 for our newly defined language-instructed ReID, +4.3% on LLCM for visible-infrared ReID, +2.6% on CUHK-PEDES for text-to-image ReID. The datasets, the model, and code are available at https://github.com/hwz-zju/Instruct-ReID.

Training deep models for LiDAR semantic segmentation is challenging due to the inherent sparsity of point clouds. Utilizing temporal data is a natural remedy against the spar-sity problem as it makes the input signal denser. However, previous multi-frame fusion algorithms fall short in utilizing sufficient temporal information due to the memory constraint, and they also ignore the informative temporal images. To fully exploit rich information hidden in long-term temporal point clouds and images, we present the Temporal Aggre-gation Network, termed TASeg. Specifically, we propose a Temporal LiDAR Aggregation and Distillation (TLAD) algorithm, which leverages historical priors to assign dif-ferent aggregation steps for different classes. It can largely reduce memory and time overhead while achieving higher accuracy. Besides, TLAD trains a teacher injected with gt priors to distill the model, further boosting the performance. To make full use of temporal images, we design a Temporal Image Aggregation and Fusion (TIAF) module, which can greatly expand the camera FOVand enhance the present features. Temporal LiDAR points in the camera FOV are used as mediums to transform temporal image features to the present coordinate for temporal multi-modal fusion. Moreover, we develop a Static-Moving Switch Augmentation (SMSA) algorithm, which utilizes sufficient temporal information to enable objects to switch their motion states freely, thus greatly increasing static and moving training samples. Our TASeg ranks 1st††the date of CVPR deadline, i.e., 2023-11-18 07:59 AM UTC. on three challenging tracks, i.e., SemanticKITTI single-scan track, multi-scan track and nuScenes LiDAR segmentation track, strongly demonstrating the superiority of our method. Codes are available at https://github.com/LittlePey/TASeg.

In the context of autonomous driving, the significance of effective feature learning is widely acknowledged. While conventional 3D self-supervised pretraining methods have shown widespread success, most methods follow the ideas originally designed for 2D images. In this paper, we present UniPAD, a novel self-supervised learning paradigm applying 3D volumetric differentiable rendering. UniPAD implicitly encodes 3D space, facilitating the reconstruction of continuous 3D shape structures and the intricate appear-ance characteristics of their 2D projections. The flexibil-ity of our method enables seamless integration into both 2D and 3D frameworks, enabling a more holistic compre-hension of the scenes. We manifest the feasibility and effectiveness of UniPAD by conducting extensive experiments on various 3D perception tasks. Our method significantly improves lidar-, camera-, and lidar-camera-based baseline by 9.1, 7.7, and 6.9 NDS, respectively. Notably, our pretraining pipeline achieves 73.2 NDS for 3D object detection and 79.4 mIoU for 3D semantic segmentation on the nuScenes validation set, achieving state-of-the-art results in comparison with previous methods.

Generative models, e.g., Stable Diffusion, have enabled the creation of photorealistic images from text prompts. Yet, the generation of 360-degree panorama images from text remains a challenge, particularly due to the dearth of paired text-panorama data and the domain gap between panorama and perspective images. In this paper, we introduce a novel dual-branch diffusion model named PanFusion to generate a 360-degree image from a text prompt. We leverage the stable diffusion model as one branch to provide prior knowledge in natural image generation and register it to another panorama branch for holistic image generation. We propose a unique cross-attention mechanism with projection awareness to minimize distortion during the collaborative denoising process. Our experiments validate that PanFusion surpasses existing methods and, thanks to its dual-branch structure, can integrate additional constraints like room layout for customized panorama outputs.

Pre-training a model and then fine-tuning it on down-stream tasks has demonstrated significant success in the 2D image and NLP domains. However, due to the unordered and non-uniform density characteristics of point clouds, it is non-trivial to explore the prior knowledge of point clouds and pre-train a point cloud backbone. In this paper, we propose a novel pre-training method called Point cloud Diffusion pre-training (PointDif). We consider the point cloud pre-training task as a conditional point-to-point gen-eration problem and introduce a conditional point genera-tor. This generator aggregates the features extracted by the backbone and employs them as the condition to guide the point-to-point recovery from the noisy point cloud, thereby assisting the backbone in capturing both local and global geometric priors as well as the global point density distri-bution of the object. We also present a recurrent uniform sampling optimization strategy, which enables the model to uniformly recover from various noise levels and learn from balanced supervision. Our PointDif achieves substan-tial improvement across various real-world datasets for di-verse downstream tasks such as classification, segmentation and detection. Specifically, PointDif attains 70.0% mIoU on S3DIS Area 5 for the segmentation task and achieves an average improvement of 2.4% on ScanObjectNN for the classification task compared to TAP. Furthermore, our pre-training framework can be flexibly applied to diverse point cloud backbones and bring considerable gains. Code is available at https://github.com/zhengxiaozx/PointDif.

Recent studies have demonstrated Large Language Models (LLMs) can extend their zero-shot generalization capabilities to multimodal learning through instruction tuning. As more modalities and downstream tasks are introduced, negative conflicts and interference may have a worse impact on performance. While this phenomenon has been overlooked in previous work, we propose a novel and extensible framework, called Octavius, for comprehensive studies and experimentation on multimodal learning with Multimodal Large Language Models (MLLMs). Specifically, to mitigate the interference, we combine the concept of Mixture-of-Experts (MoE) with LoRA and design a multimodal LoRA-MoE decoder for task- and modality-specific learning. To the best of our knowledge, we are one of the pioneering efforts to introduce MoE into MLLMs to address this problem. The experimental results (about 20% improvement) have shown the effectiveness and versatility of our design in various 2D and 3D downstream tasks. Code and corresponding dataset will be available soon.

The success of pre-trained 2D vision models can largely be attributed to their ability to learn from large-scale datasets. However, compared with 2D image datasets, current pre-training data for 3D point clouds are limited. In this paper, we propose a knowledge distillation method for pre-training 3D point cloud models by directly acquiring knowledge from a 2D representation learning model, specifically the image encoder of CLIP. To close the significant domain gap between 2D images and 3D point clouds, we propose to align the features from the two domains at the concept level. Our method utilizes a cross-attention mechanism to extract concept features from 3D point clouds and compares them with the corresponding information from 2D images. This approach bridges the two domains and allows point cloud models to learn directly from the rich information contained in 2D teacher models. Extensive experiments show that our proposed knowledge distillation scheme achieves higher accuracy than the state-of-the-art 3D pre-training methods for synthetic and real-world datasets on various downstream tasks, including object classification, object detection, semantic segmentation and part segmentation.

Precipitation nowcasting based on radar data plays a crucial role in extreme weather prediction and has broad implications for disaster management. Despite progresses have been made based on deep learning, two key challenges of precipitation nowcasting are not well-solved: (i) the modeling of complex precipitation system evolutions with different scales, and (ii) accurate forecasts for extreme precipitation. In this work, we propose CasCast, a cascaded framework composed of a deterministic and a probabilistic part to decouple the predictions for mesoscale precipitation distributions and small-scale patterns. Then, we explore training the cascaded framework at the high resolution and conducting the probabilistic modeling in a low dimensional latent space with a frame-wise-guided diffusion transformer for enhancing the optimization of extreme events while reducing computational costs. Extensive experiments on three benchmark radar precipitation datasets show that CasCast achieves competitive performance. Especially, CasCast significantly surpasses the baseline (up to +91.8%) for regional extreme-precipitation nowcasting.

In the context of this reality, existing diffusion models, such as Diffusion Transformers, often face challenges when processing image resolutions outside of their trained domain. To overcome this limitation, we present the Flexible Vision Transformer (FiT), a transformer architecture specifically designed for generating images with unrestricted resolutions and aspect ratios. Unlike traditional methods that perceive images as static-resolution grids, FiT conceptualizes images as sequences of dynamically-sized tokens. This perspective enables a flexible training strategy that effortlessly adapts to diverse aspect ratios during both training and inference phases, thus promoting resolution generalization and eliminating biases induced by image cropping. Enhanced by a meticulously adjusted network structure and the integration of training-free extrapolation techniques, FiT exhibits remarkable flexibility in resolution extrapolation generation. Comprehensive experiments demonstrate the exceptional performance of FiT across a broad range of resolutions. Repository available at https://github.com/whlzy/FiT.

Data-driven weather forecasting models are advancing rapidly, yet they rely on initial states (i.e., analysis states) typically produced by traditional data assimilation algorithms. Four-dimensional variational assimilation (4DVar) is one of the most widely adopted data assimilation algorithms in numerical weather prediction centers; it is accurate but computationally expensive. In this paper, we aim to couple the AI forecasting model, FengWu, with 4DVar to build a self-contained data-driven global weather forecasting framework, FengWu-4DVar. To achieve this, we propose an *AI-embedded* 4DVar algorithm that includes three components: (1) a 4DVar objective function embedded with the FengWu forecasting model and its error representation to enhance efficiency and accuracy; (2) a spherical-harmonic-transform-based (SHT-based) approximation strategy for capturing the horizontal correlation of background error; and (3) an auto-differentiation (AD) scheme for determining the optimal analysis fields. Experimental results show that under the ERA5 simulated observational data with varying proportions and noise levels, FengWu-4DVar can generate accurate analysis fields; remarkably, it has achieved stable self-contained global weather forecasts for an entire year for the first time, demonstrating its potential for real-world applications. Additionally, our framework is approximately 100 times faster than the traditional 4DVar algorithm under similar experimental conditions, highlighting its significant computational efficiency.

Data assimilation is a vital component in modern global medium-range weather forecasting systems to obtain the best estimation of the atmospheric state by combining the short-term forecast and observations. Recently, AI-based data assimilation approaches have attracted increasing attention for their significant advantages over traditional techniques in terms of computational consumption. However, existing AI-based data assimilation methods can only handle observations with a specific resolution, lacking the compatibility and generalization ability to assimilate observations with other resolutions. Considering that complex real-world observations often have different resolutions, we propose the Fourier Neural Processes (FNP) for arbitrary-resolution data assimilation in this paper. Leveraging the efficiency of the designed modules and flexible structure of neural processes, FNP achieves state-of-the-art results in assimilating observations with varying resolutions, and also exhibits increasing advantages over the counterparts as the resolution and the amount of observations increase. Moreover, our FNP trained on a fixed resolution can directly handle the assimilation of observations with out-of-distribution resolutions and the observational information reconstruction task without additional fine-tuning, demonstrating its excellent generalization ability across data resolutions as well as across tasks. Code is available at https://github.com/OpenEarthLab/FNP.

The success of pretrain-finetune paradigm brings about the release of numerous model weights. In this case, merging models finetuned on different tasks to enable a single model with multi-task capabilities is gaining increasing attention for its practicability. Existing model merging methods usually suffer from (1) significant performance degradation or (2) requiring tuning by additional data or training. In this paper, we rethink and analyze the existing model merging paradigm. We discover that using a single model's weights can hardly simulate all the models' performance. To tackle this issue, we propose Elect, Mask & Rescale-Merging (EMR-Merging). We first (a) elect a unified model from all the model weights and then (b) generate extremely lightweight task-specific modulators, including masks and rescalers, to align the direction and magnitude between the unified model and each specific model, respectively. EMR-Merging is tuning-free, thus requiring no data availability or any additional training while showing impressive performance. We find that EMR-Merging shows outstanding performance compared to existing merging methods under different classical and newly-established settings, including merging different numbers of vision models (up to 30), NLP models, PEFT models, and multi-modal models.

Protein language models (PLMs) have shown remarkable capabilities in various protein function prediction tasks. However, while protein function is intricately tied to structure, most existing PLMs do not incorporate protein structure information. To address this issue, we introduce ProSST, a Transformer-based protein language model that seamlessly integrates both protein sequences and structures. ProSST incorporates a structure quantization module and a Transformer architecture with disentangled attention. The structure quantization module translates a 3D protein structure into a sequence of discrete tokens by first serializing the protein structure into residue-level local structures and then embeds them into dense vector space. These vectors are then quantized into discrete structure tokens by a pre-trained clustering model. These tokens serve as an effective protein structure representation. Furthermore, ProSST explicitly learns the relationship between protein residue token sequences and structure token sequences through the sequence-structure disentangled attention. We pre-train ProSST on millions of protein structures using a masked language model objective, enabling it to learn comprehensive contextual representations of proteins. To evaluate the proposed ProSST, we conduct extensive experiments on the zero-shot mutation effect prediction and several supervised downstream tasks, where ProSST achieves the state-of-the-art performance among all baselines. Our code and pre-trained models are publicly available.

Data-driven generative models have emerged as promising approaches towards achieving efficient mechanical inverse design. However, due to prohibitively high cost in time and money, there is still lack of open-source and large-scale benchmarks in this field. It is mainly the case for airfoil inverse design, which requires to generate and edit diverse geometric-qualified and aerodynamic-qualified airfoils following the multimodal instructions, \emph{i.e.,} dragging points and physical parameters. This paper presents the open-source endeavors in airfoil inverse design, \emph{AFBench}, including a large-scale dataset with 200 thousand airfoils and high-quality aerodynamic and geometric labels, two novel and practical airfoil inverse design tasks, \emph{i.e.,} conditional generation on multimodal physical parameters, controllable editing, and comprehensive metrics to evaluate various existing airfoil inverse design methods. Our aim is to establish \emph{AFBench} as an ecosystem for training and evaluating airfoil inverse design methods, with a specific focus on data-driven controllable inverse design models by multimodal instructions capable of bridging the gap between ideas and execution, the academic research and industrial applications. We have provided baseline models, comprehensive experimental observations, and analysis to accelerate future research. Our baseline model is trained on an RTX 3090 GPU within 16 hours. The codebase, datasets and benchmarks will be available at \url{https://hitcslj.github.io/afbench/}.

Foundation models have made significant strides in understanding the genomic language of DNA sequences. However, previous models typically adopt the tokenization methods designed for natural language, which are unsuitable for DNA sequences due to their unique characteristics. In addition, the optimal approach to tokenize DNA remains largely under-explored, and may not be intuitively understood by humans even if discovered. To address these challenges, we introduce MxDNA, a novel framework where the model autonomously learns an effective DNA tokenization strategy through gradient decent. MxDNA employs a sparse Mixture of Convolution Experts coupled with a deformable convolution to model the tokenization process, with the discontinuous, overlapping, and ambiguous nature of meaningful genomic segments explicitly considered. On Nucleotide Transformer Benchmarks and Genomic Benchmarks, MxDNA demonstrates superior performance to existing methods with less pretraining data and time, highlighting its effectiveness. Finally, we show that MxDNA learns unique tokenization strategy distinct to those of previous methods and captures genomic functionalities at a token level during self-supervised pretraining. Our MxDNA aims to provide a new perspective on DNA tokenization, potentially offering broad applications in various domains and yielding profound insights. Code is available at https://github.com/qiaoqiaoLF/MxDNA.

RNA plays a pivotal role in translating genetic instructions into functional outcomes, underscoring its importance in biological processes and disease mechanisms. Despite the emergence of numerous deep learning approaches for RNA, particularly universal RNA language models, there remains a significant lack of standardized benchmarks to assess the effectiveness of these methods. In this study, we introduce the first comprehensive RNA benchmark BEACON BEnchmArk for COmprehensive RNA Task and Language Models).First, BEACON comprises 13 distinct tasks derived from extensive previous work covering structural analysis, functional studies, and engineering applications, enabling a comprehensive assessment of the performance of methods on various RNA understanding tasks. Second, we examine a range of models, including traditional approaches like CNNs, as well as advanced RNA foundation models based on language models, offering valuable insights into the task-specific performances of these models. Third, we investigate the vital RNA language model components from the tokenizer and positional encoding aspects. Notably, our findings emphasize the superiority of single nucleotide tokenization and the effectiveness of Attention with Linear Biases (ALiBi) over traditional positional encoding methods. Based on these insights, a simple yet strong baseline called BEACON-B is proposed, which can achieve outstanding performance with limited data and computational resources. The datasets and source code of our benchmark are available at https://github.com/terry-r123/RNABenchmark.

Signed Distance Function (SDF)-based volume rendering has demonstrated significant capabilities in surface reconstruction. Although promising, SDF-based methods often fail to capture detailed geometric structures, resulting in visible defects. By comparing SDF-based volume rendering to density-based volume rendering, we identify two main factors within the SDF-based approach that degrade surface quality: SDF-to-density representation and geometric regularization. These factors introduce challenges that hinder the optimization of the SDF field. To address these issues, we introduce NeuRodin, a novel two-stage neural surface reconstruction framework that not only achieves high-fidelity surface reconstruction but also retains the flexible optimization characteristics of density-based methods. NeuRodin incorporates innovative strategies that facilitate transformation of arbitrary topologies and reduce artifacts associated with density bias. Extensive evaluations on the Tanks and Temples and ScanNet++ datasets demonstrate the superiority of NeuRodin, showing strong reconstruction capabilities for both indoor and outdoor environments using solely posed RGB captures. Project website: https://open3dvlab.github.io/NeuRodin/

Data-driven artificial intelligence (AI) models have made significant advancements in weather forecasting, particularly in medium-range and nowcasting. However, most data-driven weather forecasting models are black-box systems that focus on learning data mapping rather than fine-grained physical evolution in the time dimension. Consequently, the limitations in the temporal scale of datasets prevent these models from forecasting at finer time scales. This paper proposes a physics-AI hybrid model (i.e., WeatherGFT) which generalizes weather forecasts to finer-grained temporal scales beyond training dataset. Specifically, we employ a carefully designed PDE kernel to simulate physical evolution on a small time scale (e.g., 300 seconds) and use a parallel neural networks with a learnable router for bias correction. Furthermore, we introduce a lead time-aware training framework to promote the generalization of the model at different lead times. The weight analysis of physics-AI modules indicates that physics conducts major evolution while AI performs corrections adaptively. Extensive experiments show that WeatherGFT trained on an hourly dataset, effectively generalizes forecasts across multiple time scales, including 30-minute, which is even smaller than the dataset's temporal resolution.

*Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)?* The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the **Dense Connector** - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Building on this, we also propose the Efficient Dense Connector, which achieves performance comparable to LLaVA-v1.5 with only 25% of the visual tokens. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B→70B), and diverse architectures of MLLMs (e.g., LLaVA-v1.5, LLaVA-NeXT and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development. Code is available at https://github.com/HJYao00/DenseConnector.

Diffusion-based methods have achieved remarkable achievements in 2D image or 3D object generation, however, the generation of 3D scenes and even $360^{\circ}$ images remains constrained, due to the limited number of scene datasets, the complexity of 3D scenes themselves, and the difficulty of generating consistent multi-view images. To address these issues, we first establish a large-scale panoramic video-text dataset containing millions of consecutive panoramic keyframes with corresponding panoramic depths, camera poses, and text descriptions. Then, we propose a novel text-driven panoramic generation framework, termed DiffPano, to achieve scalable, consistent, and diverse panoramic scene generation. Specifically, benefiting from the powerful generative capabilities of stable diffusion, we fine-tune a single-view text-to-panorama diffusion model with LoRA on the established panoramic video-text dataset. We further design a spherical epipolar-aware multi-view diffusion model to ensure the multi-view consistency of the generated panoramic images. Extensive experiments demonstrate that DiffPano can generate scalable, consistent, and diverse panoramic images with given unseen text descriptions and camera poses.

Large language models (LLMs) have demonstrated remarkable efficacy across knowledge-intensive tasks. Nevertheless, their untapped potential in crop science presents an opportunity for advancement. To narrow this gap, we introduce CROP, which includes a novel instruction tuning dataset specifically designed to enhance LLMs’ professional capabilities in the crop science sector, along with a benchmark that serves as a comprehensive evaluation of LLMs’ understanding of the domain knowledge. The CROP dataset is curated through a task-oriented and LLM-human integrated pipeline, comprising 210,038 single-turn and 1,871 multi-turn dialogues related to crop science scenarios. The CROP benchmark includes 5,045 multiple-choice questions covering three difficulty levels. Our experiments based on the CROP benchmark demonstrate notable enhancements in crop science-related tasks when LLMs are fine-tuned with the CROP dataset. To the best of our knowledge, CROP dataset is the first-ever instruction tuning dataset in the crop science domain. We anticipate that CROP will accelerate the adoption of LLMs in the domain of crop science, ultimately contributing to global food production.

In robot learning, the observation space is crucial due to the distinct characteristics of different modalities, which can potentially become a bottleneck alongside policy design. In this study, we explore the influence of various observation spaces on robot learning, focusing on three predominant modalities: RGB, RGB-D, and point cloud. We introduce OBSBench, a benchmark comprising two simulators and 125 tasks, along with standardized pipelines for various encoders and policy baselines. Extensive experiments on diverse contact-rich manipulation tasks reveal a notable trend: point cloud-based methods, even those with the simplest designs, frequently outperform their RGB and RGB-D counterparts. This trend persists in both scenarios: training from scratch and utilizing pre-training. Furthermore, our findings demonstrate that point cloud observations often yield better policy performance and significantly stronger generalization capabilities across various geometric and visual conditions. These outcomes suggest that the 3D point cloud is a valuable observation modality for intricate robotic tasks. We also suggest that incorporating both appearance and coordinate information can enhance the performance of point cloud methods. We hope our work provides valuable insights and guidance for designing more generalizable and robust robotic models.

Lumina-T2X is a nascent family of Flow-based Large Diffusion Transformers (Flag-DiT) that establishes a unified framework for transforming noise into various modalities, such as images and videos, conditioned on text instructions. Despite its promising capabilities, Lumina-T2X still encounters challenges including training instability, slow inference, and extrapolation artifacts. In this paper, we present Lumina-Next, an improved version of Lumina-T2X, showcasing stronger generation performance with increased training and inference efficiency. We begin with a comprehensive analysis of the Flag-DiT architecture and identify several suboptimal components, which we address by introducing the Next-DiT architecture with 3D RoPE and sandwich normalizations. To enable better resolution extrapolation, we thoroughly compare different context extrapolation methods applied to text-to-image generation with 3D RoPE, and propose Frequency- and Time-Aware Scaled RoPE tailored for diffusion transformers. Additionally, we introduce a sigmoid time discretization schedule for diffusion sampling, which achieves high-quality generation in 5-10 steps combined with higher-order ODE solvers. Thanks to these improvements, Lumina-Next not only improves the basic text-to-image generation but also demonstrates superior resolution extrapolation capabilities as well as multilingual generation using decoder-based LLMs as the text encoder, all in a zero-shot manner. To further validate Lumina-Next as a versatile generative framework, we instantiate it on diverse tasks including visual recognition, multi-views, audio, music, and point cloud generation, showcasing strong performance across these domains. By releasing all codes and model weights at https://github.com/Alpha-VLLM/Lumina-T2X, we aim to advance the development of next-generation generative AI capable of universal modeling.

In this paper, we present a groundbreaking largescale system technology that leverages optimization on global, node, and device levels to achieve unprecedented scalability for tensor networks. Our techniques enable accommodating largescale tensor networks with up to tens of terabytes of memory, reaching up to 2304 GPUs with a peak computing power of 561 PFLOPS. Notably, we have achieved a time-to-solution of 14.22 seconds with an energy consumption of 2.39 kWh which achieved a fidelity of 0.002. Our most remarkable result is a time-to-solution of 17.18 seconds, with energy consumption of only 0.29 kWh which achieved a XEB of 0.002 after post-processing. The experiments conducted demonstrate that our research outperforms Google’s quantum processor Sycamore in both speed and energy efficiency, which recorded 600 seconds and 4.3 kWh, respectively. The code is available at https://github.com/DeepLinkorg/OpenTenNet.

Accurately measuring the evolution of Multiple Sclerosis (MS) with magnetic resonance imaging (MRI) critically informs understanding of disease progression and helps to direct therapeutic strategy. Deep learning models have shown promise for automatically segmenting MS lesions, but the scarcity of accurately annotated data hinders progress in this area. Obtaining sufficient data from a single clinical site is challenging and does not address the heterogeneous need for model robustness. Conversely, the collection of data from multiple sites introduces data privacy concerns and potential label noise due to varying annotation standards. To address this dilemma, we explore the use of the federated learning framework while considering label noise. Our approach enables collaboration among multiple clinical sites without compromising data privacy under a federated learning paradigm that incorporates a noise-robust training strategy based on label correction. Specifically, we introduce a Decoupled Hard Label Correction (DHLC) strategy that considers the imbalanced distribution and fuzzy boundaries of MS lesions, enabling the correction of false annotations based on prediction confidence. We also introduce a Centrally Enhanced Label Correction (CELC) strategy, which leverages the aggregated central model as a correction teacher for all sites, enhancing the reliability of the correction process. Extensive experiments conducted on two multi-site datasets demonstrate the effectiveness and robustness of our proposed methods, indicating their potential for clinical applications in multi-site collaborations to train better deep learning models with lower cost in data collection and annotation.

Abstract Transferring knowledge from pre-trained deep models for downstream tasks, particularly with limited labeled samples, is a fundamental problem in computer vision research. Recent advances in large-scale, task-agnostic vision-language pre-trained models, which are learned with billions of samples, have shed new light on this problem. In this study, we investigate how to efficiently transfer aligned visual and textual knowledge for downstream visual recognition tasks. We first revisit the role of the linear classifier in the vanilla transfer learning framework, and then propose a new paradigm where the parameters of the classifier are initialized with semantic targets from the textual encoder and remain fixed during optimization. To provide a comparison, we also initialize the classifier with knowledge from various resources. In the empirical study, we demonstrate that our paradigm improves the performance and training speed of transfer learning tasks. With only minor modifications, our approach proves effective across 17 visual datasets that span three different data domains: image, video, and 3D point cloud.

Offline reinforcement learning (RL) methods learn from datasets without further environment interaction, facing errors due to out-of-distribution (OOD) actions. Although effective methods have been proposed to conservatively estimate the Q-values of those OOD actions to mitigate this problem, insufficient or excessive pessimism under constant constraints often harms the policy learning process. Moreover, since the distribution of each task on the dataset varies among different environments and behavior policies, it is desirable to learn an adaptive weight for balancing constraints on the conservative estimation of Q-value and the standard RL objectives depending on each task. To achieve this, in this paper, we point out that the quantile of the Q-value is an effective metric to refer to the Q-value distribution of the fixed data set. Based on this observation, we design Adaptive Pessimism via a Target Q-value (APTQ) algorithm that balances between the pessimism constraint and the RL objective; this leads the expectation of Q-value to stably converge to a given target Q-value from a reasonable quantile of the Q-value distribution of the dataset. Experiments show that our method remarkably improves the performance of the state-of-the-art method CQL by 6.20% on the D4RL-v0 and 1.89% on the D4RL-v2.

Recently, perception task based on Bird's-Eye View (BEV) representation has drawn more and more attention, and BEV representation is promising as the foundation for next-generation Autonomous Vehicle (AV) perception. However, most existing BEV solutions either require considerable resources to execute on-vehicle inference or suffer from modest performance. This paper proposes a simple yet effective framework, termed Fast-BEV, which is capable of performing faster BEV perception on the on-vehicle chips. Towards this goal, we first empirically find that the BEV representation can be sufficiently powerful without expensive transformer based transformation or depth representation. Our Fast-BEV consists of five parts, we innovatively propose (1) a lightweight deployment-friendly view transformation which fast transfers 2D image features to 3D voxel space, (2) a multi-scale image encoder which leverages multi-scale information for better performance, (3) an efficient BEV encoder which is particularly designed to speed up on-vehicle inference. We further introduce (4) a strong data augmentation strategy for both image and BEV space to avoid over-fitting, (5) a multi-frame feature fusion mechanism to leverage the temporal information. Among them, (1) and (3) enable Fast-BEV to be fast inference and deployment friendly on the on-vehicle chips, (2), (4) and (5) ensure that Fast-BEV has competitive performance. All these make Fast-BEV a solution with high performance, fast inference speed, and deployment-friendly on the on-vehicle chips of autonomous driving. Through experiments, on 2080Ti platform, our R50 model can run 52.6 FPS with 47.3% NDS on the nuScenes validation set, exceeding the 41.3 FPS and 47.5% NDS of the BEVDepth-R50 model (Li et al. 2022) and 30.2 FPS and 45.7% NDS of the BEVDet4D-R50 model (J. Huang and G. Huang, 2022). Our largest model (R101@900×1600) establishes a competitive 53.5% NDS on the nuScenes validation set. We further develop a benchmark with considerable accuracy and efficiency on current popular on-vehicle chips.

Undoubtedly, Deep Neural Networks (DNNs), from AlexNet to ResNet to Transformer, have sparked revolutionary advancements in diverse computer vision tasks. The scale of DNNs has grown exponentially due to the rapid development of computational resources. Despite the tremendous success, DNNs typically depend on massive amounts of training data (especially the recent various foundation models) to achieve high performance and are brittle in that their performance can degrade severely with small changes in their operating environment. Generally, collecting massive-scale training datasets is costly or even infeasible, as for certain fields, only very limited or no examples at all can be gathered. Nevertheless, collecting, labeling, and vetting massive amounts of practical training data is certainly difficult and expensive, as it requires the painstaking efforts of experienced human annotators or experts, and in many cases, prohibitively costly or impossible due to some reason, such as privacy, safety or ethic issues.

3D object detection from images, one of the fundamental and challenging problems in autonomous driving, has received increasing attention from both industry and academia in recent years. Benefiting from the rapid development of deep learning technologies, image-based 3D detection has achieved remarkable progress. Particularly, more than 200 works have studied this problem from 2015 to 2021, encompassing a broad spectrum of theories, algorithms, and applications. However, to date no recent survey exists to collect and organize this knowledge. In this paper, we fill this gap in the literature and provide the first comprehensive survey of this novel and continuously growing research field, summarizing the most commonly used pipelines for image-based 3D detection and deeply analyzing each of their components. Additionally, we also propose two new taxonomies to organize the state-of-the-art methods into different categories, with the intent of providing a more systematic review of existing methods and facilitating fair comparisons with future works. In retrospect of what has been achieved so far, we also analyze the current challenges in the field and discuss future directions for image-based 3D detection research.

Fine-grained image retrieval mainly focuses on learning salient features from the seen subcategories as discriminative embedding while neglecting the problems behind zero-shot settings. We argue that retrieving fine-grained objects from unseen subcategories may rely on more diverse clues, which are easily restrained by the salient features learnt from seen subcategories. To address this issue, we propose a novel Content-aware Rectified Activation model, which enables this model to suppress the activation on salient regions while preserving their discrimination, and spread activation to adjacent non-salient regions, thus mining more diverse discriminative features for retrieving unseen subcategories. Specifically, we construct a content-aware rectified prototype (CARP) by perceiving semantics of salient regions. CARP acts as a channel-wise non-destructive activation upper bound and can be selectively used to suppress salient regions for obtaining the rectified features. Moreover, two regularizations are proposed: 1) a semantic coherency constraint that imposes a restriction on semantic coherency of CARP and salient regions, aiming at propagating the discriminative ability of salient regions to CARP, 2) a feature-navigated constraint to further guide the model to adaptively balance the discrimination power of rectified features and the suppression power of salient features. Experimental results on fine-grained and product retrieval benchmarks demonstrate that our method consistently outperforms the state-of-the-art methods.

This paper proposes a novel transformer-based framework to generate accurate class-specific object localization maps for weakly supervised semantic segmentation (WSSS). Leveraging the insight that the attended regions of the one-class token in the standard vision transformer can generate class-agnostic localization maps, we investigate the transformer's capacity to capture class-specific attention for class-discriminative object localization by learning multiple class tokens. We present the Multi-Class Token transformer, which incorporates multiple class tokens to enable class-aware interactions with patch tokens. This is facilitated by a class-aware training strategy that establishes a one-to-one correspondence between output class tokens and ground-truth class labels. We also introduce a Contrastive-Class-Token (CCT) module to enhance the learning of discriminative class tokens, enabling the model to better capture the unique characteristics of each class. Consequently, the proposed framework effectively generates class-discriminative object localization maps from the class-to-patch attentions associated with different class tokens. To refine these localization maps, we propose the utilization of patch-level pairwise affinity derived from the patch-to-patch transformer attention. Furthermore, the proposed framework seamlessly complements the Class Activation Mapping (CAM) method, yielding significant improvements in WSSS performance on PASCAL VOC 2012 and MS COCO 2014. These results underline the importance of the class token for WSSS.

Transformers are widely used in computer vision areas and have achieved remarkable success. Most state-of-the-art approaches split images into regular grids and represent each grid region with a vision token. However, fixed token distribution disregards the semantic meaning of different image regions, resulting in sub-optimal performance. To address this issue, we propose the Token Clustering Transformer (TCFormer), which generates dynamic vision tokens based on semantic meaning. Our dynamic tokens possess two crucial characteristics: (1) Representing image regions with similar semantic meanings using the same vision token, even if those regions are not adjacent, and (2) concentrating on regions with valuable details and represent them using fine tokens. Through extensive experimentation across various applications, including image classification, human pose estimation, semantic segmentation, and object detection, we demonstrate the effectiveness of our TCFormer.

There is a growing need to explore the potential of transformers in Unsupervised Domain Adaptation (UDA) due to their increasing success in various vision tasks. However, the application of transformers in UDA has yet to be thoroughly investigated and requires further research. In this study, our primary focus is to design a novel pipeline specifically tailored for transformer-based UDA, to address a crucial challenge: the overemphasis on the transfer of target-oriented information, mainly caused by the self-attention blocks in transformers and the cross-domain adversarial learning scheme. First, we show that non-target information, including semantic contextual information such as background features and non-target classes, must be addressed in the domain adaptation process. Recognizing the importance of incorporating non-target knowledge, we propose a decoupled non-target knowledge distillation method called DeNKD. DeNKD decouples non-target information across domains at both feature and logit levels. This decoupling is achieved through a bi-directional knowledge distillation approach that facilitates the interaction and exchange of non-target knowledge to facilitate an effective transformer-based cross-domain knowledge transfer. We perform extensive evaluations on several well-established UDA benchmark datasets. The results consistently show that DeNKD outperforms other methods, achieving the best performance across the board. For example, on the Office-Home dataset, DeNKD achieves an accuracy of 85.54%, while on the VisDA-2017 dataset, it achieves an accuracy of 89.95%. These results highlight the effectiveness of DeNKD in transformer-based UDA and its potential for improving cross-domain adaptation performance.

Accurately retrieving surface meteorological states at arbitrary locations is of great application significance in weather forecasting and climate modeling. Since meteorological variables are typically provided as coarse-resolution gridded fields, common methods that obtain the states at a specific location directly through spatial interpolation can lead to significant accuracy deviations compared to actual observations. Traditional downscaling, the process of obtaining fixed-scale high-resolution meteorological fields from low-resolution inputs, has been proposed as a way to indirectly improve the accuracy of retrieving states at arbitrary locations by providing more detailed subgrid-scale information. However, for arbitrary locations at the station scale, their states are influenced by subgrid information, resulting in systematic biases between the downscaled results after interpolation and the actual observations at specific station locations. To address this issue, in this article, we propose a new task called station-scale downscaling, which aims to directly derive accurate meteorological states at any given station location from a coarse-resolution meteorological field. To achieve this, we propose a new downscaling model based on hypernetwork architecture, namely, HyperDS, which efficiently integrates the multiscale observational information to guide the continuous neural field modeling of the meteorological variables, enabling accurate sampling of the states at any target location. Through extensive experiments, our proposed method outperforms other specially designed baseline models on multiple surface variables. Notably, the mean squared error (mse) for wind speed and surface pressure improved by 67% and 19.5% compared with other methods, respectively.

In an era of frequent extreme weather and global warming, obtaining precise, fine-grained near-surface weather forecasts is increasingly essential for human activities. Downscaling (DS), a crucial task in meteorological forecasting and remote sensing, enables the reconstruction of high-resolution meteorological states for target regions from global-scale forecast results. Previous downscaling methods, inspired by convolutional neural network (CNN) and Transformer-based super-resolution (SR) models, lacked tailored designs for meteorology and encountered structural limitations. Notably, they failed to efficiently integrate topography, a crucial prior to the downscaling process. In this article, we address these limitations by pioneering the selective state-space model (SSM) into the meteorological field downscaling and propose a novel model called MambaDS. This model retains the advantages of Mamba in long-range dependency modeling and linear computational complexity while enhancing the learning ability of multivariate correlation. In addition, by designing an efficient topography constraint layer, this prior information can be used more efficiently than ever before. Through extensive experiments in both China mainland and the continental United States (CONUS), we validated that our proposed MambaDS achieves state-of-the-art (SOTA) results in three different types of meteorological field downscaling settings.

Context modeling is critical for remote sensing image dense prediction tasks. Nowadays, the growing size of very-high-resolution (VHR) remote sensing images poses challenges in effectively modeling context. While transformer-based models possess global modeling capabilities, they encounter computational challenges when applied to large VHR images due to their quadratic complexity. The conventional practice of cropping large images into smaller patches results in a notable loss of contextual information. To address these issues, we propose the remote sensing Mamba (RSM) for dense prediction tasks in large VHR remote sensing images. RSM is specifically designed to capture the global context of remote sensing images with linear complexity, facilitating the effective processing of large VHR images. Considering that the land covers in remote sensing images are distributed in arbitrary spatial directions due to characteristics of remote sensing over-head imaging, the RSM incorporates an omnidirectional selective scan module (OSSM) to globally model the context of images in multiple directions, capturing large spatial features from various directions. We designed simple yet effective models based on RSM, achieving state-of-the-art performance on dense prediction tasks in VHR remote sensing images without fancy training strategies. Extensive experiments on semantic segmentation (SS) and change detection (CD) tasks across various land covers demonstrate the effectiveness of the proposed RSM. Leveraging the linear complexity and global modeling capabilities, RSM achieves better efficiency and accuracy than transformer-based models on large remote sensing images. Interestingly, we also demonstrated that our model generally performs better with a larger image size on dense prediction tasks.

Nonintrusive load monitoring (NILM) can monitor the operating state and energy consumption of electric appliances in a nonintrusive manner and provides a promising approach to improving electricity usage efficiency for residential and commercial buildings. Although machine learning (ML) methods are powerful and have significantly advanced the developments of NILM, they request a sizable amount of labeled data for model training. However, getting operational data of each electrical appliance in real life is challenging, so the requirements for labeled data limit the NLIM's practicality. To tackle this challenge, a novel multilayer momentum contrast (MLMoCo) learning mechanism is proposed for self-supervised feature representation learning. With only unlabeled aggregate load data, the proposed MLMoCo contrasts the augmented versions of the same sample (“positives”) with instances extracted from other samples (“negatives”). To maintain a dictionary with enough negative samples to be compared with the input, a momentum encoder is adopted to momentum update the parameters rather than by backpropagation during training. An event-based data augmentation method is also proposed to obtain the distinct but strongly related positive pairs for self-supervised feature learning. The experimental comparisons, including different state-of-the-art techniques and various downstream tasks with real-world datasets, demonstrate the remarkable performance gains of the proposed approach through learning from the unlabeled data, which could significantly increase the practicality of the NILM.

As one of the classic problems of pattern recognition, the online Handwritten Chinese Character Recognition (OLHCCR) has attracted the attention of many researchers. Yet, it remains challenging due to complex glyphs, numerous strokes, and huge categories. Existing methods utilize temporal features or spatial features to recognize handwritten characters, which results in recognition errors due to the character with non-standard stroke order. This paper proposes a new OLHCCR model based on 1-D Convolution and Two-Streams Transformers. The model has a 1-D Transformer and a Vision Transformer, and the 1-D Transformer contains a 1-D Convolution layer and Transformers, that is, the model has overall structure of Two-Streams Transformers with 1-D Convolution. So, the model is named as C-TST. It can fuse temporal and spatial features of Chinese character to achieve high recognition accuracy and fast recognition speed. Specifically, each online handwritten Chinese character is represented by a trajectory sequence. The original trajectory sequence is preprocessed to enhance the information density of each trajectory point and features difference among trajectory points. Then, the result after preprocessing is input into the 1-D convolution layer to obtain shallow temporal features, which are used also as the input of the Transformers to capture the temporal features. Simultaneously, character image is generated by processing the original trajectory sequence, and then fed into the Vision Transformer to capture the spatial features. By fusing the captured temporal and spatial features of online handwritten Chinese character, the proposed C-TST achieves a recognition accuracy of 97.90% on ICDAR-2013 and a state-of-the-art recognition accuracy of 97.38% on IAHCC-UCAS2016. The code will be available athttps://github.com/cwnuiot/Two-Streams-Transformer

Joint detection and tracking, which solves two fundamental vision challenges in a unified manner, is a challenging topic in computer vision. In this area, the proper use of spatial-temporal information in videos can help reduce local defects and improve the quality of feature representations. Although modeling low-level (usually pixel-wise) spatial-temporal information has been studied, instance-level spatial-temporal correlations (i.e., relations between semantic regions in which instances have occurred) have not been fully exploited. In comparison, modeling instance-level correlation is a more flexible and reasonable way to enhance feature representations. However, we have found that conventional instance-level relation learning that works for the separate tasks of detection or tracking is not effective in joint tasks in which a variety of scenarios may be presented. To try to resolve this problem, in this study, we effectively exploited instance-level spatial-temporal semantic information for joint detection and tracking via a joint relation learning pipeline with a novel relation learning mechanism called Similarity- and Quality-Guided Attention (SQGA). Specifically, we added task-specific SQGA relation modules before the corresponding task prediction heads to refine the instance feature representation using features of other reference instances in the neighboring frames; these features are aggregated on the basis of relational affinities. In particular, in SQGA, relational affinities were factorized to similarity and quality terms so that fine-grained supervision rules could be applied. Then we added task-specific attention losses for each SQGA relation module, resulting in a better feature aggregation for the corresponding task. Quantitative experiments based on several challenging multi-object tracking benchmarks showed that our approach was more effective than the baselines and provided competitive results compared with recent state-of-the-art methods.

Observing that the existing model compression approaches only focus on reducing the redundancies in convolutional neural networks (CNNs) along one particular dimension (e.g., the channel or spatial or temporal dimension), in this work, we propose our multidimensional pruning (MDP) framework, which can compress both 2-D CNNs and 3-D CNNs along multiple dimensions in an end-to-end fashion. Specifically, MDP indicates the simultaneous reduction of channels and more redundancy on other additional dimensions. The redundancy of additional dimensions depends on the input data, i.e., spatial dimension for 2-D CNNs when using images as the input data, and spatial and temporal dimensions for 3-D CNNs when using videos as the input data. We further extend our MDP framework to the MDP-Point approach for compressing point cloud neural networks (PCNNs) whose inputs are irregular point clouds (e.g., PointNet). In this case, the redundancy along the additional dimension indicates the point dimension (i.e., the number of points). Comprehensive experiments on six benchmark datasets demonstrate the effectiveness of our MDP framework and its extended version MDP-Point for compressing CNNs and PCNNs, respectively.

Multi-agent reinforcement learning (MARL) suffers from the non-stationarity problem, which is the ever-changing targets at every iteration when multiple agents update their policies at the same time. Starting from first principle, in this paper, we manage to solve the non-stationarity problem by proposing bidirectional action-dependent Q-learning (ACE). Central to the development of ACE is the sequential decision making process wherein only one agent is allowed to take action at one time. Within this process, each agent maximizes its value function given the actions taken by the preceding agents at the inference stage. In the learning phase, each agent minimizes the TD error that is dependent on how the subsequent agents have reacted to their chosen action. Given the design of bidirectional dependency, ACE effectively turns a multi-agent MDP into a single-agent MDP. We implement the ACE framework by identifying the proper network representation to formulate the action dependency, so that the sequential decision process is computed implicitly in one forward pass. To validate ACE, we compare it with strong baselines on two MARL benchmarks. Empirical experiments demonstrate that ACE outperforms the state-of-the-art algorithms on Google Research Football and StarCraft Multi-Agent Challenge by a large margin. In particular, on SMAC tasks, ACE achieves 100% success rate on almost all the hard and super hard maps. We further study extensive research problems regarding ACE, including extension, generalization and practicability.

Fine-grained object retrieval aims to learn discriminative representation to retrieve visually similar objects. However, existing top-performing works usually impose pairwise similarities on the semantic embedding spaces or design a localization sub-network to continually fine-tune the entire model in limited data scenarios, thus resulting in convergence to suboptimal solutions. In this paper, we develop Fine-grained Retrieval Prompt Tuning (FRPT), which steers a frozen pre-trained model to perform the fine-grained retrieval task from the perspectives of sample prompting and feature adaptation. Specifically, FRPT only needs to learn fewer parameters in the prompt and adaptation instead of fine-tuning the entire model, thus solving the issue of convergence to suboptimal solutions caused by fine-tuning the entire model. Technically, a discriminative perturbation prompt (DPP) is introduced and deemed as a sample prompting process, which amplifies and even exaggerates some discriminative elements contributing to category prediction via a content-aware inhomogeneous sampling operation. In this way, DPP can make the fine-grained retrieval task aided by the perturbation prompts close to the solved task during the original pre-training. Thereby, it preserves the generalization and discrimination of representation extracted from input samples. Besides, a category-specific awareness head is proposed and regarded as feature adaptation, which removes the species discrepancies in features extracted by the pre-trained model using category-guided instance normalization. And thus, it makes the optimized features only include the discrepancies among subcategories. Extensive experiments demonstrate that our FRPT with fewer learnable parameters achieves the state-of-the-art performance on three widely-used fine-grained datasets.

Synthesizing controllable motion for a character using deep learning has been a promising approach due to its potential to learn a compact model without laborious feature engineering. To produce dynamic motion from weak control signals such as desired paths, existing methods often require auxiliary information such as phases for alleviating motion ambiguity, which limits their generalisation capability. As past poses often contain useful auxiliary hints, in this paper, we propose a task-agnostic deep learning method, namely Multi-scale Control Signal-aware Transformer (MCS-T), with an attention based encoder-decoder architecture to discover the auxiliary information implicitly for synthesizing controllable motion without explicitly requiring auxiliary information such as phase. Specifically, an encoder is devised to adaptively formulate the motion patterns of a character's past poses with multi-scale skeletons, and a decoder driven by control signals to further synthesize and predict the character's state by paying context-specialised attention to the encoded past motion patterns. As a result, it helps alleviate the issues of low responsiveness and slow transition which often happen in conventional methods not using auxiliary information. Both qualitative and quantitative experimental results on an existing biped locomotion dataset, which involves diverse types of motion transitions, demonstrate the effectiveness of our method. In particular, MCS-T is able to successfully generate motions comparable to those generated by the methods using auxiliary information.

Transferring knowledge from task-agnostic pre-trained deep models for downstream tasks is an important topic in computer vision research. Along with the growth of computational capacity, we now have open-source vision-language pre-trained models in large scales of the model architecture and amount of data. In this study, we focus on transferring knowledge for video classification tasks. Conventional methods randomly initialize the linear classifier head for vision classification, but they leave the usage of the text encoder for downstream visual recognition tasks undiscovered. In this paper, we revise the role of the linear classifier and replace the classifier with the different knowledge from pre-trained model. We utilize the well-pretrained language model to generate good semantic target for efficient transferring learning. The empirical study shows that our method improves both the performance and the training speed of video classification, with a negligible change in the model. Our simple yet effective tuning paradigm achieves state-of-the-art performance and efficient training on various video recognition scenarios, i.e., zero-shot, few-shot, general recognition. In particular, our paradigm achieves the state-of-the-art accuracy of 87.8% on Kinetics-400, and also surpasses previous methods by 20~50% absolute top-1 accuracy under zero-shot, few-shot settings on five video datasets. Code and models are available at https://github.com/whwu95/Text4Vis.

LiDAR and Radar are two complementary sensing approaches in that LiDAR specializes in capturing an object's 3D shape while Radar provides longer detection ranges as well as velocity hints. Though seemingly natural, how to efficiently combine them for improved feature representation is still unclear. The main challenge arises from that Radar data are extremely sparse and lack height information. Therefore, directly integrating Radar features into LiDAR-centric detection networks is not optimal. In this work, we introduce a bi-directional LiDAR-Radar fusion framework, termed Bi-LRFusion, to tackle the challenges and improve 3D detection for dynamic objects. Technically, Bi-LRFusion involves two steps: first, it enriches Radar's local features by learning important details from the LiDAR branch to alleviate the problems caused by the absence of height information and extreme sparsity; second, it combines LiDAR features with the enhanced Radar features in a unified bird's-eye-view representation. We conduct extensive experiments on nuScenes and ORR datasets, and show that our Bi-LRFusion achieves state-of-the-art performance for detecting dynamic objects. Notably, Radar data in these two datasets have different formats, which demonstrates the generalizability of our method. Codes will be published.

Human-centric perceptions (e.g., pose estimation, human parsing, pedestrian detection, person re-identification, etc.) play a key role in industrial applications of visual models. While specific human-centric tasks have their own relevant semantic aspect to focus on, they also share the same underlying semantic structure of the human body. However, few works have attempted to exploit such homogeneity and design a general-propose model for human-centric tasks. In this work, we revisit a broad range of human-centric tasks and unify them in a minimalist manner. We propose UniHCP, a Unified Model for Human-Centric Perceptions, which unifies a wide range of human-centric tasks in a simplified end-to-end manner with the plain vision transformer architecture. With large-scale joint training on 33 human-centric datasets, UniHCP can outperform strong baselines on several in-domain and downstream tasks by direct evaluation. When adapted to a specific task, UniHCP achieves new SOTAs on a wide range of human-centric tasks, e.g., 69.8 mIoU on CIHP for human parsing, 86.18 mA on PA100K for attribute prediction, 90.3 mAP on Market1501 for ReID, and 85.8 JI on CrowdHuman for pedestrian detection, performing better than specialized models tailored for each task. The code and pretrained model are available at https://github.com/OpenGVLab/UniHCP.

Existing image captioning methods are under the assumption that the training and testing data are from the same domain or that the data from the target domain (i.e., the domain that testing data lie in) are accessible. However, this assumption is invalid in real-world applications where the data from the target domain is inaccessible. In this paper, we introduce a new setting called Domain Generalization for Image Captioning (DGIC), where the data from the target domain is unseen in the learning process. We first construct a benchmark dataset for DGIC, which helps us to investigate models' domain generalization (DG) ability on unseen domains. With the support of the new benchmark, we further propose a new framework called language-guided semantic metric learning (LSML) for the DGIC setting. Experiments on multiple datasets demonstrate the challenge of the task and the effectiveness of our newly proposed benchmark and LSML framework.

Human-centric perceptions include a variety of vision tasks, which have widespread industrial applications, including surveillance, autonomous driving, and the metaverse. It is desirable to have a general pretrain model for versatile human-centric downstream tasks. This paper forges ahead along this path from the aspects of both benchmark and pretraining methods. Specifically, we propose a HumanBench based on existing datasets to comprehensively evaluate on the common ground the generalization abilities of different pretraining methods on 19 datasets from 6 diverse downstream tasks, including person ReID, pose estimation, human parsing, pedestrian attribute recognition, pedestrian detection, and crowd counting. To learn both coarse-grained and fine-grained knowledge in human bodies, we further propose a Projector AssisTed Hierarchical pretraining method (PATH) to learn diverse knowledge at different granularity levels. Comprehensive evaluations on HumanBench show that our PATH achieves new state-of-the-art results on 17 downstream datasets and on-par results on the other 2 datasets. The code will be publicly at https://github.com/OpenGVLab/HumanBench.

Open-set fine-grained retrieval is an emerging challenge that requires an extra capability to retrieve unknown subcategories during evaluation. However, current works focus on close-set visual concepts, where all the subcategories are pre-defined, and make it hard to capture discriminative knowledge from unknown subcategories, consequently failing to handle unknown subcategories in open-world scenarios. In this work, we propose a novel Prompting vision-Language Evaluator (PLEor) framework based on the recently introduced contrastive language-image pretraining (CLIP) model, for open-set fine-grained retrieval. PLEor could leverage pre-trained CLIP model to infer the discrepancies encompassing both pre-defined and unknown subcategories, called category-specific discrepancies, and transfer them to the backbone network trained in the close-set scenarios. To make pre-trained CLIP model sensitive to category-specific discrepancies, we design a dual prompt scheme to learn a vision prompt specifying the categoryspecific discrepancies, and turn random vectors with category names in a text prompt into category-specific discrepancy descriptions. Moreover, a vision-language evaluator is proposed to semantically align the vision and text prompts based on CLIP model, and reinforce each other. In addition, we propose an open-set knowledge transfer to transfer the category-specific discrepancies into the backbone network using knowledge distillation mechanism. Quantitative and qualitative experiments show that our PLEor achieves promising performance on open-set fine-grained datasets.

Most existing text-video retrieval methods focus on cross-modal matching between the visual content of videos and textual query sentences. However, in real-world scenarios, online videos are often accompanied by relevant text information such as titles, tags, and even subtitles, which can be utilized to match textual queries. This insight has motivated us to propose a novel approach to text-video retrieval, where we directly generate associated captions from videos using zero-shot video captioning with knowledge from web-scale pre-trained models (e.g., CLIP and GPT-2). Given the generated captions, a natural question arises: what benefits do they bring to text-video retrieval? To answer this, we introduce Cap4Video, a new framework that leverages captions in three ways: i) Input data: video-caption pairs can augment the training data. ii) Intermediate feature interaction: we perform cross-modal feature interaction between the video and caption to produce enhanced video representations. iii) Output score: the Query-Caption matching branch can complement the original Query-Video matching branch for text-video retrieval. We conduct comprehensive ablation studies to demonstrate the effectiveness of our approach. Without any post-processing, Cap4Video achieves state-of-the-art performance on four standard text-video retrieval benchmarks: MSR-VTT (51.4%), VATEX (66.6%), MSVD (51.8%), and DiDeMo (52.0%). The code is available at https://github.com/whwu95/Cap4Video.

Vision-language models (VLMs) pre-trained on large- scale image-text pairs have demonstrated impressive transferability on various visual tasks. Transferring knowledge from such powerful VLMs is a promising direction for building effective video recognition models. However, current exploration in this field is still limited. We believe that the greatest value of pre-trained VLMs lies in building a bridge between visual and textual domains. In this paper, we propose a novel framework called BIKE, which utilizes the cross-modal bridge to explore bidirectional knowledge: i) We introduce the Video Attribute Association mechanism, which leverages the Video-to-Text knowledge to generate textual auxiliary attributes for complementing video recognition. ii) We also present a Temporal Concept Spotting mechanism that uses the Text-to-Video expertise to capture temporal saliency in a parameter-free manner, leading to enhanced video representation. Extensive studies on six popular video datasets, including Kinetics-400 & 600, UCF-101, HMDB-51, ActivityNet and Charades, show that our method achieves state-of-the-art performance in various recognition scenarios, such as general, zero-shot, and few-shot video recognition. Our best model achieves a state-of-the-art accuracy of 88.6% on the challenging Kinetics-400 using the released CLIP model. The code is available at https://github.com/whwu95/BIKE.

Weakly supervised dense object localization (WSDOL) relies generally on Class Activation Mapping (CAM), which exploits the correlation between the class weights of the image classifier and the pixel-level features. Due to the limited ability to address intra-class variations, the image classifier cannot properly associate the pixel features, leading to inaccurate dense localization maps. In this paper, we propose to explicitly construct multi-modal class representations by leveraging the Contrastive Language-Image Pre-training (CLIP), to guide dense localization. More specifically, we propose a unified transformer framework to learn two-modalities of class-specific tokens, i.e., class-specific visual and textual tokens. The former captures semantics from the target visual data while the latter exploits the class-related language priors from CLIP, providing complementary information to better perceive the intra-class diversities. In addition, we propose to enrich the multi-modal class-specific tokens with sample-specific contexts comprising visual context and image-language context. This enables more adaptive class representation learning, which further facilitates dense localization. Extensive experiments show the superiority of the proposed method for WSDOL on two multi-label datasets, i.e., PASCAL VOC and MS COCO, and one single-label dataset, i.e., OpenImages. Our dense localization maps also lead to the state-of-the-art weakly supervised semantic segmentation (WSSS) results on PASCAL VOC and MS COCO.11https://github.com/xulianuwa/MMCST

Masked Modeling (MM) has demonstrated widespread success in various vision challenges, by reconstructing masked visual patches. Yet, applying MM for large-scale 3D scenes remains an open problem due to the data sparsity and scene complexity. The conventional random masking paradigm used in 2D images often causes a high risk of ambiguity when recovering the masked region of 3D scenes. To this end, we propose a novel informative-preserved reconstruction, which explores local statistics to discover and preserve the representative structured points, effectively enhancing the pretext masking task for 3D scene understanding. Integrated with a progressive reconstruction manner, our method can concentrate on modeling regional geometry and enjoy less ambiguity for masked reconstruction. Besides, such scenes with progressive masking ratios can also serve to self-distill their intrinsic spatial consistency, requiring to learn the consistent representations from unmasked areas. By elegantly combining informative-preserved reconstruction on masked areas and consistency self-distillation from unmasked areas, a unified framework called MM-3DScene is yielded. We conduct comprehensive experiments on a host of downstream tasks. The consistent improvement (e.g., +6.1% [email protected] on object detection and +2.2% mIoU on semantic segmentation) demonstrates the superiority of our approach.

Recent Transformer-based 3D object detectors learn point cloud features either from point- or voxel-based representations. However, the former requires time-consuming sampling while the latter introduces quantization errors. In this paper, we present a novel Point-Voxel Transformer for single-stage 3D detection (PVT-SSD) that takes advantage of these two representations. Specifically, we first use voxel-based sparse convolutions for efficient feature encoding. Then, we propose a Point-Voxel Transformer (PVT) module that obtains long-range contexts in a cheap manner from voxels while attaining accurate positions from points. The key to associating the two different representations is our introduced input-dependent Query Initialization module, which could efficiently generate reference points and content queries. Then, PVT adaptively fuses long-range contextual and local geometric information around reference points into content queries. Further, to quickly find the neighboring points of reference points, we design the Virtual Range Image module, which generalizes the native range image to multi-sensor and multi-frame. The experiments on several autonomous driving benchmarks verify the effectiveness and efficiency of the proposed method. Code will be available.

Despite the tremendous progress of Masked Autoencoders (MAE) in developing vision tasks such as image and video, exploring MAE in large-scale 3D point clouds remains challenging due to the inherent irregularity. In contrast to previous 3D MAE frameworks, which either design a complex decoder to infer masked information from maintained regions or adopt sophisticated masking strategies, we instead propose a much simpler paradigm. The core idea is to apply a Generative Decoder for MAE (GD-MAE) to automatically merges the surrounding context to restore the masked geometric knowledge in a hierarchical fusion manner. In doing so, our approach is free from introducing the heuristic design of decoders and enjoys the flexibility of exploring various masking strategies. The corresponding part costs less than 12% latency compared with conventional methods, while achieving better performance. We demonstrate the efficacy of the proposed method on several large-scale benchmarks: Waymo, KITTI, and ONCE. Consistent improvement on downstream detection tasks illustrates strong robustness and generalization capability. Not only our method reveals state-of-the-art results, but remarkably, we achieve comparable accuracy even with 20% of the labeled data on the Waymo dataset. Code will be released.

Model-based reinforcement learning (RL) has demonstrated remarkable successes on a range of continuous control tasks due to its high sample efficiency. To save the computation cost of conducting planning online, recent practices tend to distill optimized action sequences into an RL policy during the training phase. Although the distillation can incorporate both the foresight of planning and the exploration ability of RL policies, the theoretical understanding of these methods is yet unclear. In this paper, we extend the policy improvement of Soft Actor-Critic (SAC) by developing an approach to distill from model-based planning to the policy. We then demonstrate that such an approach of policy improvement has a theoretical guarantee of monotonic improvement and convergence to the maximum value defined in SAC. We discuss effective design choices and implement our theory as a practical algorithm—Model-based Planning Distilled to Policy (MPDP)—that updates the policy jointly over multiple future time steps. Extensive experiments show that MPDP achieves better sample efficiency and asymptotic performance than both model-free and model-based planning algorithms on six continuous control benchmark tasks in MuJoCo.

Scale variation is a deep-rooted problem in object counting, which has not been effectively addressed by existing scale-aware algorithms. An important factor is that they typically involve cooperative learning across multi-resolutions, which could be suboptimal for learning the most discriminative features from each scale. In this paper, we propose a novel method termed STEERER (SelecTivE inhERitance lEaRning) that addresses the issue of scale variations in object counting. STEERER selects the most suitable scale for patch objects to boost feature extraction and only inherits discriminative features from lower to higher resolution progressively. The main insights of STEERER are a dedicated Feature Selection and Inheritance Adaptor (FSIA), which selectively forwards scale-customized features at each scale, and a Masked Selection and Inheritance Loss (MSIL) that helps to achieve high-quality density maps across all scales. Our experimental results on nine datasets with counting and localization tasks demonstrate the unprecedented scale generalization ability of STEERER. Code is available at https://github.com/taohan10200/STEERER.

Pre-training across 3D vision and language remains under development because of limited training data. Recent works attempt to transfer vision-language (V-L) pre-training methods to 3D vision. However, the domain gap between 3D and images is unsolved, so that V-L pre-trained models are restricted in 3D downstream tasks. To address this issue, we propose CLIP2Point, an image-depth pre-training method by contrastive learning to transfer CLIP to the 3D domain, and adapt it to point cloud classification. We introduce a new depth rendering setting that forms a better visual effect, and then render 52,460 pairs of images and depth maps from ShapeNet for pre-training. The pre-training scheme of CLIP2Point combines cross-modality learning to enforce the depth features for capturing expressive visual and textual features and intra-modality learning to enhance the invariance of depth aggregation. Additionally, we propose a novel Gated Dual-Path Adapter (GDPA), i.e., a dual-path structure with global-view aggregators and gated fusion for downstream representative learning. It allows the ensemble of CLIP and CLIP2Point, tuning pre-training knowledge to downstream tasks in an efficient adaptation. Experimental results show that CLIP2Point is effective in transferring CLIP knowledge to 3D vision. CLIP2Point outperforms other 3D transfer learning and pre-training networks, achieving state-of-the-art results on zero-shot, few-shot, and fully-supervised classification. Codes are available at: https://github.com/tyhuang0428/CLIP2Point.

We propose a novel approach to self-supervised learning of point cloud representations by differentiable neural rendering. Motivated by the fact that informative point cloud features should be able to encode rich geometry and appearance cues and render realistic images, we train a point-cloud encoder within a devised point-based neural renderer by comparing the rendered images with real images on massive RGB-D data. The learned point-cloud encoder can be easily integrated into various downstream tasks, including not only high-level tasks like 3D detection and segmentation but also low-level tasks like 3D reconstruction and image synthesis. Extensive experiments on various tasks demonstrate the superiority of our approach compared to existing pre-training methods.

Semi-supervised semantic segmentation methods use a small amount of clean pixel-level annotations to guide the interpretation of a larger quantity of unlabelled image data. The challenges of providing pixel-accurate annotations at scale mean that the labels are typically noisy, and this contaminates the final results. In this work, we propose an approach that is robust to label noise in the annotated data. The method uses two diverse learning groups with different network architectures to effectively handle both label noise and unlabelled images. Each learning group consists of a teacher network, a student network and a novel filter module. The filter module of each learning group utilizes pixel-level features from the teacher network to detect incorrectly labelled pixels. To reduce confirmation bias, we employ the labels cleaned by the filter module from one learning group to train the other learning group. Experimental results on two different benchmarks and settings demonstrate the superiority of our method over state-of-the-art approaches.

In this work, we build a modular-designed codebase, formulate strong training recipes, design an error diagnosis toolbox, and discuss current methods for image-based 3D object detection. In particular, different from other highly mature tasks, e.g., 2D object detection, the community of image-based 3D object detection is still evolving, where methods often adopt different training recipes and tricks resulting in unfair evaluations and comparisons. What is worse, these tricks may overwhelm their proposed designs in performance, even leading to wrong conclusions. To address this issue, we build a module-designed code-base and formulate unified training standards for the community. Furthermore, we also design an error diagnosis toolbox to measure the detailed characterization of detection models. Using these tools, we analyze current methods in-depth under varying settings and provide discussions for some open questions, e.g., discrepancies in conclusions on KITTI-3D and nuScenes datasets, which have led to different dominant methods for these datasets. We hope that this work will facilitate future research in image-based 3D object detection. Our codes will be released at https://github.com/OpenGVLab/3dodi.

In 3D human action recognition, limited supervised data makes it challenging to fully tap into the modeling potential of powerful networks such as transformers. As a result, researchers have been actively investigating effective self-supervised pre-training strategies. In this work, we show that instead of following the prevalent pretext task to perform masked self-component reconstruction in human joints, explicit contextual motion modeling is key to the success of learning effective feature representation for 3D action recognition. Formally, we propose the Masked Motion Prediction (MAMP) framework. To be specific, the proposed MAMP takes as input the masked spatio-temporal skeleton sequence and predicts the corresponding temporal motion of the masked human joints. Considering the high temporal redundancy of the skeleton sequence, in our MAMP, the motion information also acts as an empirical semantic richness prior that guide the masking process, promoting better attention to semantically rich temporal regions. Extensive experiments on NTU-60, NTU-120, and PKU-MMD datasets show that the proposed MAMP pre-training substantially improves the performance of the adopted vanilla transformer, achieving state-of-the-art results without bells and whistles. The source code of our MAMP is available at https://github.com/maoyunyao/MAMP.

Temporal modeling plays a crucial role in understanding video content. To tackle this problem, previous studies built complicated temporal relations through time sequence thanks to the development of computationally powerful devices. In this work, we explore the potential of four simple arithmetic operations for temporal modeling. Specifically, we first capture auxiliary temporal cues by computing addition, subtraction, multiplication, and division between pairs of extracted frame features. Then, we extract corresponding features from these cues to benefit the original temporal-irrespective domain. We term such a simple pipeline as an Arithmetic Temporal Module (ATM), which operates on the stem of a visual backbone with a plug-and-play style. We conduct comprehensive ablation studies on the instantiation of ATMs and demonstrate that this module provides powerful temporal modeling capability at a low computational cost. Moreover, the ATM is compatible with both CNNs- and ViTs-based architectures. Our results show that ATM achieves superior performance over several popular video benchmarks. Specifically, on Something-Something V1, V2 and Kinetics-400, we reach top-1 accuracy of 65.6%, 74.6%, and 89.4% respectively. The code is available at https://github.com/whwu95/ATM.

Monocular 3D Semantic Scene Completion (SSC) has garnered significant attention in recent years due to its potential to predict complex semantics and geometry shapes from a single image, requiring no 3D inputs. In this paper, we identify several critical issues in current state-of-the-art methods, including the Feature Ambiguity of projected 2D features in the ray to the 3D space, the Pose Ambiguity of the 3D convolution, and the Computation Imbalance in the 3D convolution across different depth levels. To address these problems, we devise a novel Normalized Device Coordinates scene completion network (NDC-Scene) that directly extends the 2D feature map to a Normalized Device Coordinates (NDC) space, rather than to the world space directly, through progressive restoration of the dimension of depth with deconvolution operations. Experiment results demonstrate that transferring the majority of computation from the target 3D space to the proposed normalized device coordinates space benefits monocular SSC tasks. Additionally, we design a Depth-Adaptive Dual Decoder to simultaneously upsample and fuse the 2D and 3D feature maps, further improving overall performance. Our extensive experiments confirm that the proposed method consistently outperforms state-of-the-art methods on both outdoor SemanticKITTI and indoor NYUv2 datasets. Our code are available at https://github.com/Jiawei-Yao0812/NDCScene.

The low-quality cost distribution obtained by simple upsampling leads to disparity maps with many outliers and low sub-pixel accuracy. We propose the Cost Distribution Super-Resolution Network (CDSRNet), which directly extracts high-resolution cost distribution from the low-resolution 4D cost volume.The similarity extraction module of CDSRNet decomposes the task of estimating high-resolution cost distribution into multiple subtasks and completes each subtask by a specific translation block, ensuring high discrimination of the predicted cost distribution. The inter-block aggregation module aggregates information from other subtasks, obtaining a voting volume with global information for correcting errors in the current subtask. Experiment results demonstrate that the proposed method significantly reduces the ratio of outliers and improves the sub-pixel accuracy.

Photos serve as a way for humans to record what they experience in their daily lives, and they are often regarded as trustworthy sources of information. However, there is a growing concern that the advancement of artificial intelligence (AI) technology may produce fake photos, which can create confusion and diminish trust in photographs. This study aims to comprehensively evaluate agents for distinguishing state-of-the-art AI-generated visual content. Our study benchmarks both human capability and cutting-edge fake image detection AI algorithms, using a newly collected large-scale fake image dataset Fake2M. In our human perception evaluation, titled HPBench, we discovered that humans struggle significantly to distinguish real photos from AI-generated ones, with a misclassification rate of 38.7\%. Along with this, we conduct the model capability of AI-Generated images detection evaluation MPBench and the top-performing model from MPBench achieves a 13\% failure rate under the same setting used in the human evaluation.We hope that our study can raise awareness of the potential risks of AI-generated images and facilitate further research to prevent the spread of false information. More information can refer to https://github.com/Inf-imagine/Sentry.

Open-set fine-grained retrieval is an emerging challenging task that allows to retrieve unknown categories beyond the training set. The best solution for handling unknown categories is to represent them using a set of visual attributes learnt from known categories, as widely used in zero-shot learning. Though important, attribute modeling usually requires significant manual annotations and thus is labor-intensive. Therefore, it is worth to investigate how to transform retrieval models trained by image-level supervision from category semantic extraction to attribute modeling. To this end, we propose a novel Visual Attribute Parameterization Network (VAPNet) to learn visual attributes from known categories and parameterize them into the retrieval model, without the involvement of any attribute annotations. In this way, VAPNet could utilize its parameters to parse a set of visual attributes from unknown categories and precisely represent them. Technically, VAPNet explicitly attains some semantics with rich details via making use of local image patches and distills the visual attributes from these discovered semantics. Additionally, it integrates the online refinement of these visual attributes into the training process to iteratively enhance their quality. Simultaneously, VAPNet treats these attributes as supervisory signals to tune the retrieval models, thereby achieving attribute parameterization. Extensive experiments on open-set fine-grained retrieval datasets validate the superior performance of our VAPNet over existing solutions.

Currently, LiDAR-based 3D detectors are broadly categorized into two groups, namely, BEV-based detectors and cluster-based detectors. BEV-based detectors capture the contextual information from the Bird's Eye View (BEV) and fill their center voxels via feature diffusion with a stack of convolution layers, which, however, weakens the capability of presenting an object with the center point. On the other hand, cluster-based detectors exploit the voting mechanism and aggregate the foreground points into object-centric clusters for further prediction. In this paper, we explore how to effectively combine these two complementary representations into a unified framework. Specifically, we propose a new 3D object detection framework, referred to as CluB, which incorporates an auxiliary cluster-based branch into the BEV-based detector by enriching the object representation at both feature and query levels. Technically, CluB is comprised of two steps. First, we construct a cluster feature diffusion module to establish the association between cluster features and BEV features in a subtle and adaptive fashion. Based on that, an imitation loss is introduced to distill object-centric knowledge from the cluster features to the BEV features. Second, we design a cluster query generation module to leverage the voting centers directly from the cluster branch, thus enriching the diversity of object queries. Meanwhile, a direction loss is employed to encourage a more accurate voting center for each cluster. Extensive experiments are conducted on Waymo and nuScenes datasets, and our CluB achieves state-of-the-art performance on both benchmarks.

Large language models have emerged as a promising approach towards achieving general-purpose AI agents. The thriving open-source LLM community has greatly accelerated the development of agents that support human-machine dialogue interaction through natural language processing. However, human interaction with the world extends beyond only text as a modality, and other modalities such as vision are also crucial. Recent works on multi-modal large language models, such as GPT-4V and Bard, have demonstrated their effectiveness in handling visual modalities. However, the transparency of these works is limited and insufficient to support academic research. To the best of our knowledge, we present one of the very first open-source endeavors in the field, LAMM, encompassing a Language-Assisted Multi-Modal instruction tuning dataset, framework, and benchmark. Our aim is to establish LAMM as a growing ecosystem for training and evaluating MLLMs, with a specific focus on facilitating AI agents capable of bridging the gap between ideas and execution, thereby enabling seamless human-AI interaction. Our main contribution is three-fold: 1) We present a comprehensive dataset and benchmark, which cover a wide range of vision tasks for 2D and 3D vision. Extensive experiments validate the effectiveness of our dataset and benchmark. 2) We outline the detailed methodology of constructing multi-modal instruction tuning datasets and benchmarks for MLLMs, enabling rapid scaling and extension of MLLM research to diverse domains, tasks, and modalities. 3) We provide a primary but potential MLLM training framework optimized for modality extension. We also provide baseline models, comprehensive experimental observations, and analysis to accelerate future research. Our baseline model is trained within 24 A100 GPU hours, framework supports training with V100 and RTX3090 is available thanks to the open-source society. Codes and data are now available at https://openlamm.github.io.

Self-supervised sound source localization in unconstrained visual scenes is an important task of audio-visual learning. In this paper, we propose a visual reasoning module to explicitly exploit the rich visual context semantics, which alleviates the issue of insufficient utilization of visual information in previous works. The learning objectives are carefully designed to provide stronger supervision signals for the extracted visual semantics while enhancing the audio-visual interactions, which lead to more robust feature representations. Extensive experimental results demonstrate that our approach significantly boosts the localization performances on various datasets, even without initializations pretrained on ImageNet. Moreover, with the visual context exploitation, our framework can accomplish both the audio-visual and purely visual inference, which expands the application scope of the sound source localization task and further raises the competitiveness of our approach.

Recent years have witnessed the success of deep learning on the visual sound separation task. However, existing works follow similar settings where the training and testing datasets share the same musical instrument categories, which to some extent limits the versatility of this task. In this work, we focus on a more general and challenging scenario, namely the separation of unknown musical instruments, where the categories in training and testing phases have no direct overlap with each other. To tackle this new setting, we propose the "Separation-with-Consistency" (SeCo) framework, which can accomplish the separation on unknown categories by exploiting the consistency constraints. Furthermore, to capture richer characteristics of the novel melodies, we devise an online matching strategy, which can bring stable enhancements with no cost of extra parameters. Experiments demonstrate that our SeCo framework exhibits strong adaptation ability on the novel musical categories and outperforms the baseline methods by a notable margin.

Abstract Weakly supervised semantic segmentation (WSSS) commonly relies on Class Activation Mapping (CAM) to produce pseudo semantic labels using image-level annotations. However, because CAM maps often form sparse object regions with poor boundaries, they cannot provide sufficient segmentation supervision. Because off-the-shelf saliency maps can provide rich object boundaries that can be leveraged to improve semantic segmentation, we propose to jointly learn semantic segmentation and class-agnostic masks by using image-level annotations and off-the-shelf saliency maps as supervision. We also propose a cross-task label refinement mechanism, which takes advantage of the learned class-agnostic masks and semantic segmentation masks, to refine the pseudo labels and provide more accurate supervision to both tasks. Moreover, we introduce a new normalization method for CAM to generate more complete class-specific localization maps. The improved CAM maps complement our learned class-agnostic masks, leading to high-quality pseudo semantic segmentation labels. Extensive experiments demonstrate the effectiveness of the proposed approach, with state-of-the-art WSSS results established on PASCAL VOC 2012 and MS COCO.

In this work, we explore neat yet effective Transformer-based frameworks for visual grounding. The previous methods generally address the core problem of visual grounding, i.e., multi-modal fusion and reasoning, with manually-designed mechanisms. Such heuristic designs are not only complicated but also make models easily overfit specific data distributions. To avoid this, we first propose TransVG, which establishes multi-modal correspondences by Transformers and localizes referred regions by directly regressing box coordinates. We empirically show that complicated fusion modules can be replaced by a simple stack of Transformer encoder layers with higher performance. However, the core fusion Transformer in TransVG is stand-alone against uni-modal encoders, and thus should be trained from scratch on limited visual grounding data, which makes it hard to be optimized and leads to sub-optimal performance. To this end, we further introduce TransVG++ to make two-fold improvements. For one thing, we upgrade our framework to a purely Transformer-based one by leveraging Vision Transformer (ViT) for vision feature encoding. For another, we devise Language Conditioned Vision Transformer that removes external fusion modules and reuses the uni-modal ViT for vision-language fusion at the intermediate layers. We conduct extensive experiments on five prevalent datasets, and report a series of state-of-the-art records.

Semantic segmentation has achieved huge progress via adopting deep Fully Convolutional Networks (FCN). However, the performance of FCN-based models severely rely on the amounts of pixel-level annotations which are expensive and time-consuming. Considering that bounding boxes also contain abundant semantic and objective information, an intuitive solution is to learn the segmentation with weak supervisions from the bounding boxes. How to make full use of the class-level and region-level supervisions from bounding boxes to estimate the uncertain regions is the critical challenge for the weakly supervised learning task. In this paper, we propose a mixture model to address this problem. First, we introduce a box-driven class-wise masking model (BCM) to remove irrelevant regions of each class. Moreover, based on the pixel-level segment proposal generated from the bounding box supervision, we calculate the mean filling rates of each class to serve as an important prior cue to guide the model ignoring the wrongly labeled pixels in proposals. To realize the more fine-grained supervision at instance-level, we further propose the anchor-based filling rate shifting module. Unlike previous methods that directly train models with the generated noisy proposals, our method can adjust the model learning dynamically with the adaptive segmentation loss. Thus it can help reduce the negative impacts from wrongly labeled proposals. Besides, based on the learned high-quality proposals with above pipeline, we explore to further boost the performance through two-stage learning. The proposed method is evaluated on the challenging PASCAL VOC 2012 benchmark and achieves 74.9 % and 76.4 % mean IoU accuracy under weakly and semi-supervised modes, respectively. Extensive experimental results show that the proposed method is effective and is on par with, or even better than current state-of-the-art methods.

Long-tail distribution is widely spread in real-world applications. Due to the extremely small ratio of instances, tail categories often show inferior accuracy. In this paper, we find such performance bottleneck is mainly caused by the imbalanced gradients, which can be categorized into two parts: (1) positive part, deriving from the samples of the same category, and (2) negative part, contributed by other categories. Based on comprehensive experiments, it is also observed that the gradient ratio of accumulated positives to negatives is a good indicator to measure how balanced a category is trained. Inspired by this, we come up with a gradient-driven training mechanism to tackle the long-tail problem: re-balancing the positive/negative gradients dynamically according to current accumulative gradients, with a unified goal of achieving balance gradient ratios. Taking advantage of the simple and flexible gradient mechanism, we introduce a new family of gradient-driven loss functions, namely equalization losses. We conduct extensive experiments on a wide spectrum of visual tasks, including two-stage/single-stage long-tailed object detection (LVIS), long-tailed image classification (ImageNet-LT, Places-LT, iNaturalist), and long-tailed semantic segmentation (ADE20 K). Our method consistently outperforms the baseline models, demonstrating the effectiveness and generalization ability of the proposed equalization losses.

This paper investigates the task of 2D whole-body human pose estimation, which aims to localize dense landmarks on the entire human body including body, feet, face, and hands. We propose a single-network approach, termed ZoomNet, to take into account the hierarchical structure of the full human body and solve the scale variation of different body parts. We further propose a neural architecture search framework, termed ZoomNAS, to promote both the accuracy and efficiency of whole-body pose estimation. ZoomNAS jointly searches the model architecture and the connections between different sub-modules, and automatically allocates computational complexity for searched sub-modules. To train and evaluate ZoomNAS, we introduce the first large-scale 2D human whole-body dataset, namely COCO-WholeBody V1.0, which annotates 133 keypoints for in-the-wild images. Extensive experiments demonstrate the effectiveness of ZoomNAS and the significance of COCO-WholeBody V1.0.

Trajectory forecasting for traffic participants (e.g., vehicles) is critical for autonomous platforms to make safe plans. Currently, most trajectory forecasting methods assume that object trajectories have been extracted and directly develop trajectory predictors based on the ground truth trajectories. However, this assumption does not hold in practical situations. Trajectories obtained from object detection and tracking are inevitably noisy, which could cause serious forecasting errors to predictors built on ground truth trajectories. In this paper, we propose to predict trajectories directly based on detection results without relying on explicitly formed trajectories. Different from traditional methods which encode the motion cues of an agent based on its clearly defined trajectory, we extract the motion information only based on the affinity cues among detection results, in which an affinity-aware state update mechanism is designed to manage the state information. In addition, considering that there could be multiple plausible matching candidates, we aggregate the states of them. These designs take the uncertainty of association into account which relax the undesirable effect of noisy trajectory obtained from data association and improve the robustness of the predictor. Extensive experiments validate the effectiveness of our method and its generalization ability to different detectors or forecasting schemes.

We present SiamSampler, the first to our knowledge investigating video sampling in visual object tracking. We observe that the random sampling applied in Siamese-based trackers cannot focus on important data or ensure data diversity, hindering the effective training of networks. This paper proposes the Video-Guided Sampling Strategy to solve the problems in random sampling from both inter and intra-video levels. At the inter-video level, we propose Modified Gaussian Sampling Strategy (MGSS) to automatically assign higher sampling probabilities to longer and more difficult videos and reduce the sampling probabilities of shorter and easier videos. At the intra-video level, the Farthest Image Pair Sampling Strategy (FPSS) is proposed to increase the diversity of training data. Extensive experiments on general benchmarks demonstrate the effectiveness of our method. Compared with the baseline model, our method improves tracking performance on five datasets, without affecting the testing speed.

Unsupervised domain adaption (UDA) aims to reduce the domain gap between labeled source and unlabeled target domains. Many prior works exploit adversarial learning that leverages pre-designed discriminators to drive the network for aligning distributions between domains. However, most of them do not consider the degeneration of the domain discriminators caused by the gradually dominating gradients of aligned target samples during training, and they still suffer from the cross-domain semantic mismatch problem in the learned feature space. Hence, this paper attempts to understand and solve both issues from the lens of optimization loss and propose an automatic loss function search for adversarial domain adaptation (ALSDA). First, we extend the common adversarial loss by adding an adjustable hyper-parameter that can re-weight the gradients assigned to target samples, so that the domain discriminator can impose consecutive and influential driving forces for domain alignment. Meanwhile, we upgrade the traditional orthogonality loss with class-wisely adjustable hyper-parameters that can strengthen the cross-domain feature separation. Since manually determining the optimal loss functions requires expensive expert efforts, we leverage the popular AutoML to automatically search for the optimal loss functions from a pre-defined novel and unique search space for UDA. Further, to enable the loss function search when the target domain is unlabeled, we introduce a simple-but-effective entropy-guided search strategy with the aid of REINFORCE learning. Extensive experiments on various typical baselines and benchmark datasets such as Office-Home, Office-31, and Birds-31 have been conducted, and the results validate the generalization and superiority of the proposed ALSDA.

Existing fine-grained image recognition works have attempted to dig into low-level details for emphasizing subtle discrepancies among sub-categories. However, a potential limitation of these methods is that they integrate the low-level details and high-level semantics directly, and neglect their content complementarity and spatial corresponding correlation. To handle this limitation, we propose an end-to-end Semantic-guided Information Alignment Network (SIA-Net) to dynamically pick out the low-level details under the guidance of accurate semantics to make selected details spatially corresponding to high-level semantics and complementary in content. Technically, SIA-Net consists of an Accurate Semantic Calibration (ASC) module for providing accurate semantics and a Discriminative Feature Alignment (DFA) module for aggregating low-level details and high-level semantics using accurate semantics generated by ASC. ASC learns the pixel-level feature shifting caused by convolutional operations, which is utilized for replacing the incorrectly highlighted semantics by shifting discriminative semantics or background features. After obtaining the accurate semantic features, DFA digs into the complementary details and simultaneously makes the selected details spatially corresponding via applying the guidance of accurate semantics to obtain the reassembly features. Finally, the reassembly features, which serve as discriminative cues, are used for more accurate discriminative region localization. Extensive experiments verify that our proposed method yields the best performance under the same settings with the most competitive approaches on CUB-birds, Stanford-Cars, and FGVC Aircraft datasets.

Group activity recognition, which aims to simultaneously understand individual action and group activity in video clips, plays a fundamental role in computer vision and video analysis. In this paper, we propose a novel relational inference framework, termed MLST-Former, for individual action and group activity recognition, capturing the various degrees of spatial-temporal dynamic interactions adaptively and jointly among actors and generating reasonable group representations. Specifically, we first design a multi-level spatial-temporal Transformer to capture the miscellaneous actors’ spatial-temporal contextual information to deal with unbalanced interactions between actors. Furthermore, our proposed network is capable of fully mining long-range spatial-temporal dependencies with the virtue of the merge function and cross attention mechanism. We then propose an inter-frame gating fusion mechanism (IGFM) to selectively aggregate the temporal and structural features of the interacting actors. A new multi-task learning strategy, consisting of the classification cost of individual actions and group activities, is also developed. Moreover, we adopt the motion trajectory branch to provide complementary dynamic features for improving recognition performance. A series of ablation studies demonstrate the effectiveness and respective contributions of the different components within the proposed method. Extensive experiments on four public GAR datasets clearly show that our approach can achieve very competitive performance by comparing them with state-of-the-art methods.

Pedestrian detection is still a challenging task for computer vision, especially in crowded scenes where the overlaps between pedestrians tend to be large. The non-maximum suppression (NMS) plays an important role in removing the redundant false positive detection proposals while retaining the true positive detection proposals. However, the highly overlapped results may be suppressed if the threshold of NMS is lower. Meanwhile, a higher threshold of NMS will introduce a larger number of false positive results. To solve this problem, we propose an optimal threshold prediction (OTP) based NMS method that predicts a suitable threshold of NMS for each human instance. First, a visibility estimation module is designed to obtain the visibility ratio. Then, we propose a threshold prediction subnet to determine the optimal threshold of NMS automatically according to the visibility ratio and classification score. Finally, we re-formulate the objective function of the subnet and utilize the reward-guided gradient estimation algorithm to update the subnet. Comprehensive experiments on CrowdHuman and CityPersons show the superior performance of the proposed method in pedestrian detection, especially in crowded scenes.

Employing additional prior knowledge to model local features as a final fine-grained object representation has become a trend for fine-grained object retrieval (FGOR). A potential limitation of these methods is that they only focus on common parts across the dataset (e.g. head, body or even leg) by introducing additional prior knowledge, but the retrieval of a fine-grained object may rely on category-specific nuances that contribute to category prediction. To handle this limitation, we propose an end-to-end Category-specific Nuance Exploration Network (CNENet) that elaborately discovers category-specific nuances that contribute to category prediction, and semantically aligns these nuances grouped by subcategory without any additional prior knowledge, to directly emphasize the discrepancy among subcategories. Specifically, we design a Nuance Modelling Module that adaptively predicts a group of category-specific response (CARE) maps via implicitly digging into category-specific nuances, specifying the locations and scales for category-specific nuances. Upon this, two nuance regularizations are proposed: 1) semantic discrete loss that forces each CARE map to attend to different spatial regions to capture diverse nuances; 2) semantic alignment loss that constructs a consistent semantic correspondence for each CARE map of the same order with the same subcategory via guaranteeing each instance and its transformed counterpart to be spatially aligned. Moreover, we propose a Nuance Expansion Module, which exploits context appearance information of discovered nuances and refines the prediction of current nuance by its similar neighbors, leading to further improvement on nuance consistency and completeness. Extensive experiments validate that our CNENet consistently yields the best performance under the same settings against most competitive approaches on CUB Birds, Stanford Cars, and FGVC Aircraft datasets.

Multiple modalities can provide rich semantic information; and exploiting such information will normally lead to better performance compared with the single-modality counterpart. However, it is not easy to devise an effective cross-modal fusion structure due to the variations of feature dimensions and semantics, especially when the inputs even come from different sensors, as in the field of audio-visual learning. In this work, we propose SepFusion, a novel framework that can smoothly produce optimal fusion structures for visual-sound separation. The framework is composed of two components, namely the model generator and the evaluator. To construct the generator, we devise a lightweight architecture space that can adapt to different input modalities. In this way, we can easily obtain audio-visual fusion structures according to our demands. For the evaluator, we adopt the idea of neural architecture search to select superior networks effectively. This automatic process can significantly save human efforts while achieving competitive performances. Moreover, since our SepFusion provides a series of strong models, we can utilize the model family for broader applications, such as further promoting performance via model assembly, or providing suitable architectures for the separation of certain instrument classes. These potential applications further enhance the competitiveness of our approach.

Pedestrian counting is a fundamental tool for under-standing pedestrian patterns and crowd flow analysis. Existing works (e.g., image-level pedestrian counting, cross-line crowd counting et al.) either only focus on the image-level counting or are constrained to the manual annotation of lines. In this work, we propose to conduct the pedes-trian counting from a new perspective - Video Individual Counting (VIC), which counts the total number of individual pedestrians in the given video (a person is only counted once). Instead of relying on the Multiple Object Tracking (MOT) techniques, we propose to solve the problem by decomposing all pedestrians into the initial pedestrians who existed in the first frame and the new pedestrians with separate identities in each following frame. Then, an end-to-end Decomposition and Reasoning Network (DRNet) is designed to predict the initial pedestrian count with the density estimation method and reason the new pedestrian's count of each frame with the differentiable optimal transport. Extensive experiments are conducted on two datasets with congested pedestrians and diverse scenes, demonstrating the effectiveness of our method over baselines with great superiority in counting the individual pedestrians. Code: https://github.com/taohan10200/DRNet.

Unsupervised learning has been popular in various computer vision tasks, including visual object tracking. However, prior unsupervised tracking approaches rely heavily on spatial supervision from templatesearch pairs and are still unable to track objects with strong variation over a long time span. As unlimited self-supervision signals can be obtained by tracking a video along a cycle in time, we investigate evolving a Siamese tracker by tracking videos forward-backward. We present a novel unsupervised tracking framework, in which we can learn temporal correspondence both on the classification branch and regression branch. Specifically, to propagate reliable template feature in the forward propagation process so that the tracker can be trained in the cycle, we first propose a consistency propagation transformation. We then identify an ill-posed penalty problem in conventional cycle training in backward propagation process. Thus, a differentiable region mask is proposed to select features as well as to implicitly penalize tracking errors on intermediate frames. Moreover, since noisy labels may degrade training, we propose a mask-guided loss reweighting strategy to assign dynamic weights based on the quality of pseudo labels. In extensive experiments, our tracker outperforms preceding unsupervised methods by a substantial margin, performing on par with supervised methods on large-scale datasets such as TrackingNet and LaSOT. Code is available at https://github.com/FlorinShum/ULAST.

The pretrain-finetune paradigm is a classical pipeline in visual learning. Recent progress on unsupervised pretraining methods shows superior transfer performance to their supervised counterparts. This paper revisits this phenomenon and sheds new light on understanding the transferability gap between unsupervised and supervised pretraining from a multilayer perceptron (MLP) perspective. While previous works [6], [8], [17] focus on the effectiveness of MLP on unsupervised image classification where pretraining and evaluation are conducted on the same dataset, we reveal that the MLP projector is also the key factor to better transferability of unsupervised pretraining methods than supervised pretraining methods. Based on this observation, we attempt to close the transferability gap between supervised and unsupervised pretraining by adding an MLP projector before the classifier in supervised pretraining. Our analysis indicates that the MLP projector can help retain intra-class variation of visual features, decrease the feature distribution distance between pretraining and evaluation datasets, and reduce feature redundancy. Extensive experiments on public benchmarks demonstrate that the added MLP projector significantly boosts the transferability of supervised pretraining, e.g. +7.2% top-1 accuracy on the concept generalization task, +5.8% top-1 accuracy for linear evaluation on 12 -domain classification tasks, and +0.8% AP on COCO object detection task, making supervised pretraining comparable or even better than unsupervised pretraining.

This paper proposes a new transformer-based framework to learn class-specific object localization maps as pseudo labels for weakly supervised semantic segmentation (WSSS). Inspired by the fact that the attended regions of the one-class token in the standard vision transformer can be leveraged to form a class-agnostic localization map, we investigate if the transformer model can also effectively capture class-specific attention for more discriminative object localization by learning multiple class tokens within the transformer. To this end, we propose a Multi-class Token Transformer, termed as MCTformer, which uses multiple class tokens to learn interactions between the class tokens and the patch tokens. The proposed MCTformer can successfully produce class-discriminative object localization maps from the class-to-patch attentions corresponding to different class tokens. We also propose to use a patch-level pairwise affinity, which is extracted from the patch-to-patch transformer attention, to further refine the localization maps. Moreover, the proposed framework is shown to fully complement the Class Activation Mapping (CAM) method, leading to remarkably superior WSSS results on the PASCAL VOC and MS COCO datasets. These results underline the importance of the class token for WSSS.11https://github.com/xulianuwa/MCTformer

Neural Architecture Search (NAS) has attracted increasingly more attention in recent years because of its capability to design deep neural network automatically. Among them, differential NAS approaches such as DARTS, have gained popularity for the search efficiency. However, they suffer from two main issues, the weak robustness to the performance collapse and the poor generalization ability of the searched architectures. To solve these two problems, a simple-but-efficient regularization method, termed as Beta-Decay, is proposed to regularize the DARTS-based NAS searching process. Specifically, Beta-Decay regularization can impose constraints to keep the value and variance of activated architecture parameters from too large. Furthermore, we provide in-depth theoretical analysis on how it works and why it works. Experimental results on NAS-Bench-201 show that our proposed method can help to stabilize the searching process and makes the searched network more transferable across different datasets. In addition, our search scheme shows an outstanding property of being less dependent on training time and data. Comprehensive experiments on a variety of search spaces and datasets validate the effectiveness of the proposed method. The code is available at https://github.com/Sunshine-Ye/Beta-DARTS.

Vision transformers have achieved great successes in many computer vision tasks. Most methods generate vision tokens by splitting an image into a regular and fixed grid and treating each cell as a token. However, not all regions are equally important in human-centric vision tasks, e.g., the human body needs a fine representation with many tokens, while the image background can be modeled by a few tokens. To address this problem, we propose a novel Vision Transformer, called Token Clustering Transformer (TCFormer), which merges tokens by progressive clustering, where the tokens can be merged from different locations with flexible shapes and sizes. The tokens in TCFormer can not only focus on important areas but also adjust the token shapes to fit the semantic concept and adopt a fine resolution for regions containing critical details, which is beneficial to capturing detailed information. Extensive experiments show that TCFormer consistently outperforms its counterparts on different challenging human-centric tasks and datasets, including whole-body pose estimation on COCO-WholeBody and 3D human mesh reconstruction on 3DPW. Code is available at https://github.com/zengwang430521/TCFormer.git.

Recently, self-supervised vision transformers have attracted unprecedented attention for their impressive representation learning ability. However, the dominant method, contrastive learning, mainly relies on an instance discrimination pretext task, which learns a global understanding of the image. This paper incorporates local feature learning into self-supervised vision transformers via Reconstructive Pre-training (RePre). Our RePre extends contrastive frameworks by adding a branch for reconstructing raw image pixels in parallel with the existing contrastive objective. RePre equips with a lightweight convolution-based decoder that fuses the multi-hierarchy features from the transformer encoder. The multi-hierarchy features provide rich supervisions from low to high semantic information, crucial for our RePre. Our RePre brings decent improvements on various contrastive frameworks with different vision transformer architectures. Transfer performance in downstream tasks outperforms supervised pre-training and state-of-the-art (SOTA) self-supervised counterparts.

Unsupervised pretraining methods for object detection aim to learn object discrimination and localization ability from large amounts of images. Typically, recent works design pretext tasks that supervise the detector to predict the defined object priors. They normally leverage heuristic methods to produce object priors, \emph{e.g.,} selective search, which separates the prior generation and detector learning and leads to sub-optimal solutions. In this work, we propose a novel object detection pretraining framework that could generate object priors and learn detectors jointly by generating accurate object priors from the model itself. Specifically, region priors are extracted by attention maps from the encoder, which highlights foregrounds. Instance priors are the selected high-quality output bounding boxes of the detection decoder. By assuming objects as instances in the foreground, we can generate object priors with both region and instance priors. Moreover, our object priors are jointly refined along with the detector optimization. With better object priors as supervision, the model could achieve better detection capability, which in turn promotes the object priors generation. Our method improves the competitive approaches by \textbf{+1.3 AP}, \textbf{+1.7 AP} in 1\% and 10\% COCO low-data regimes object detection.

Residual networks have shown great success and become indispensable in today’s deep models. In this work, we aim to re-investigate the training process of residual networks from a novel social psychology perspective of loafing, and further propose a new training strategy to strengthen the performance of residual networks. As residual networks can be viewed as ensembles of relatively shallow networks (i.e., unraveled view) in prior works, we also start from such view and consider that the final performance of a residual network is co-determined by a group of sub-networks. Inspired by the social loafing problem of social psychology, we find that residual networks invariably suffer from similar problem, where sub-networks in a residual network are prone to exert less effort when working as part of the group compared to working alone. We define this previously overlooked problem as network loafing. As social loafing will ultimately cause the low individual productivity and the reduced overall performance, network loafing will also hinder the performance of a given residual network and its sub-networks. Referring to the solutions of social psychology, we propose stimulative training, which randomly samples a residual sub-network and calculates the KL-divergence loss between the sampled sub-network and the given residual network, to act as extra supervision for sub-networks and make the overall goal consistent. Comprehensive empirical results and theoretical analyses verify that stimulative training can well handle the loafing problem, and improve the performance of a residual network by improving the performance of its sub-networks. The code is available at https://github.com/Sunshine-Ye/NIPS22-ST.

This paper presents an approach that reconstructs a hand-held object from a monocular video. In contrast to many recent methods that directly predict object geometry by a trained network, the proposed approach does not require any learned prior about the object and is able to recover more accurate and detailed object geometry. The key idea is that the hand motion naturally provides multiple views of the object and the motion can be reliably estimated by a hand pose tracker. Then, the object geometry can be recovered by solving a multi-view reconstruction problem. We devise an implicit neural representation-based method to solve the reconstruction problem and address the issues of imprecise hand pose estimation, relative hand-object motion, and insufficient geometry optimization for small objects. We also provide a newly collected dataset with 3D ground truth to validate the proposed approach. The dataset and code will be released at https://dihuangdh.github.io/hhor.

As a challenging task of high-level video understanding, weakly supervised temporal action localization has attracted more attention recently. With only video-level category labels, this task should indistinguishably identify the background and action categories frame by frame. However, it is non-trivial to achieve this in untrimmed videos, due to the unconstrained background, complex and multi-label actions. With the observation that these difficulties are mainly brought by the large variations within background and actions, we propose to address these challenges from the perspective of modeling variations. Moreover, it is desired to further reduce the variations, or learn compact features, so as to cast the problem of background identification as rejecting background and alleviate the contradiction between classification and detection. Accordingly, in this paper, we propose a two-branch relational prototypical network. The first branch, namely action-branch, adopts class-wise prototypes and mainly acts as an auxiliary to introduce priori knowledge about label dependencies and be a guide for the second branch. Meanwhile, the second branch, namely sub-branch, starts with multiple prototypes, namely sub-prototypes, to enable a powerful ability of modeling variations. As a further benefit, we elaborately design a multi-label clustering loss based on the sub-prototypes to learn compact features under the multi-label setting. The two branches are associated using the correspondences between two types of prototypes, leading to a special two-stage classifier in the s-branch, on the other hand, the two branches serve as regularization terms to each other, improving the final performance. Ablation studies find that the proposed model is capable of modeling classes with large variations and learning compact features. Extensive experimental evaluations on Thumos14, MultiThumos and ActivityNet datasets demonstrate the effectiveness of the proposed method and superior performance over state-of-the-art approaches.

Multi-scale representations deeply learned via convolutional neural networks have shown tremendous importance for various pixel-level prediction problems. In this paper we present a novel approach that advances the state of the art on pixel-level prediction in a fundamental aspect, i.e. structured multi-scale features learning and fusion. In contrast to previous works directly considering multi-scale feature maps obtained from the inner layers of a primary CNN architecture, and simply fusing the features with weighted averaging or concatenation, we propose a probabilistic graph attention network structure based on a novel Attention-Gated Conditional Random Fields (AG-CRFs) model for learning and fusing multi-scale representations in a principled manner. In order to further improve the learning capacity of the network structure, we propose to exploit feature dependant conditional kernels within the deep probabilistic framework. Extensive experiments are conducted on four publicly available datasets (i.e. BSDS500, NYUD-V2, KITTI and Pascal-Context) and on three challenging pixel-wise prediction problems involving both discrete and continuous labels (i.e. monocular depth estimation, object contour prediction and semantic segmentation). Quantitative and qualitative results demonstrate the effectiveness of the proposed latent AG-CRF model and the overall probabilistic graph attention network with feature conditional kernels for structured feature learning and pixel-wise prediction.

Reconstructing 3D human shape and pose from monocular images is challenging despite the promising results achieved by the most recent learning-based methods. The commonly occurred misalignment comes from the facts that the mapping from images to the model space is highly non-linear and the rotation-based pose representation of the body model is prone to result in the drift of joint positions. In this work, we investigate learning 3D human shape and pose from dense correspondences of body parts and propose a Decompose-and-aggregate Network (DaNet) to address these issues. DaNet adopts the dense correspondence maps, which densely build a bridge between 2D pixels and 3D vertexes, as intermediate representations to facilitate the learning of 2D-to-3D mapping. The prediction modules of DaNet are decomposed into one global stream and multiple local streams to enable global and fine-grained perceptions for the shape and pose predictions, respectively. Messages from local streams are further aggregated to enhance the robust prediction of the rotation-based poses, where a position-aided rotation feature refinement strategy is proposed to exploit spatial relationships between body joints. Moreover, a Part-based Dropout (PartDrop) strategy is introduced to drop out dense information from intermediate representations during training, encouraging the network to focus on more complementary body parts as well as neighboring position features. The efficacy of the proposed method is validated on both indoor and real-world datasets including Human3.6M, UP3D, COCO, and 3DPW, showing that our method could significantly improve the reconstruction performance in comparison with previous state-of-the-art methods. Our code is publicly available at https://hongwenzhang.github.io/dense2mesh.

In the task of pedestrian trajectory prediction, social interaction could be one of the most complicated factors since it is difficult to be interpreted through simple rules. Recent studies have shown a great ability of LSTM networks in learning social behaviors from datasets, e.g., introducing LSTM hidden states of the neighbors at the last time step into LSTM recursion. However, those methods depend on previous neighboring features which lead to a delayed observation. In this paper, we propose a data-driven states refinement LSTM network (SR-LSTM) to enable the utilization of the current intention of neighbors through a message passing framework. Moreover, the model performs in the form of self-updating by jointly refining the current states of all participants, rather than an input-output mechanism served by feature concatenation. In the process of states refinement, a social-aware information selection module consisting of an element-wise motion gate and a pedestrian-wise attention is designed to serve as the guidance of the message passing process. Considering the pedestrian walking space as a graph where each pedestrian is a node and each pedestrian pair with an edge, spatial-edge LSTMs are further exploited to enhance the model capacity, where two kinds of LSTMs interact with each other so that states of them are interactively refined. Experimental results on four widely used pedestrian trajectory datasets, ETH, UCY, PWPD, and NYGC demonstrate the effectiveness of the proposed model.

The strong demand of autonomous driving in the industry has led to vigorous interest in 3D object detection and resulted in many excellent 3D object detection algorithms. However, the vast majority of algorithms only model single-frame data, ignoring the temporal clue in video sequence. In this work, we propose a new transformer, called Temporal-Channel Transformer (TCTR), to model the temporal-channel domain and spatial-wise relationships for video object detecting from Lidar data. As the special design of this transformer, the information encoded in the encoder is different from that in the decoder. The encoder encodes temporal-channel information of multiple frames while the decoder decodes the spatial-wise information for the current frame in a voxel-wise manner. Specifically, the temporal-channel encoder of the transformer is designed to encode the information of different channels and frames by utilizing the correlation among features from different channels and frames. On the other hand, the spatial decoder of the transformer decodes the information for each location of the current frame. Before conducting the object detection with detection head, a gate mechanism is further deployed for re-calibrating the features of current frame, which filters out the object-irrelevant information by repetitively refining the representation of target frame along with the up-sampling process. Experimental results reveal that TCTR achieves the state-of-the-art performance in grid voxel-based 3D object detection on the nuScenes benchmark.

Existing stereo matching methods achieve satisfactory average accuracy on a whole predicted disparity map under common global metrics, but ignore the fine-grained performance at region level, especially for far regions in the scene, which is more crucial in actual auto-driving scenarios. There are two factors accounting for this problem: 1) Depth resolution. Existing methods use disparity-based sampling to extract matching candidates uniformly according to the disparity range, but leads to sparser sampling density at far regions than that of close regions in terms of depth range, resulting in low depth resolution in far regions. 2) Feature discriminability. Limited image resolution and inferior feature extraction at far regions result in the obtained features with low discrimination, which influences the subsequent matching process between stereo images. To improve the estimation accuracy of far regions and thus achieve a balanced performance at region level, we design a novel two-stage Balanced Stereo Matching Network (BSMNet) to address the above problems. The coarse stage of BSMNet introduces a direct depth-based sampling strategy, which generates matching candidates according to scene depth instead of disparity, thus improving the depth resolution and obtaining initial depth map with more balanced accuracy. Then, a depth refinement stage is proposed to solve the problem of low feature discriminability and further optimizes the initial depth map obtained from the coarse stage. It selects matching candidates and computes their similarity scores from a carefully designed adaptive feature volume guided by a learnable scale map, thus making the final estimation more accurate. Different from the existing methods that construct stereo matching based on disparity prediction, our proposed pipeline is to directly optimize on depth information. Experiments show that our BSMNet can obtain an obvious performance improvement at far regions without discarding that at close regions, so as to largely outperform existing state-of-the-art methods.

Motion modeling is crucial in modern action recognition methods. As motion dynamics like moving tempos and action amplitude may vary a lot in different video clips, it poses great challenge on adaptively covering proper motion information. To address this issue, we introduce a Motion Diversification and Selection (MoDS) module to generate diversified spatio-temporal motion features and then select the suitable motion representation dynamically for categorizing the input video. To be specific, we first propose a spatio-temporal motion generation (StMG) module to construct a bank of diversified motion features with varying spatial neighborhood and time range. Then, a dynamic motion selection (DMS) module is leveraged to choose the most discriminative motion feature both spatially and temporally from the feature bank. As a result, our proposed method can make full use of the diversified spatio-temporal motion information, while maintaining computational efficiency at the inference stage. Extensive experiments on five widely-used benchmarks, demonstrate the effectiveness of the method and we achieve state-of-the-art performance on Something-Something V1 & V2 that are of large motion variation.

Human beings can quickly adapt to environmental changes by leveraging learning experience. However, adapting deep neural networks to dynamic environments by machine learning algorithms remains a challenge. To better understand this issue, we study the problem of continual domain adaptation, where the model is presented with a labelled source domain and a sequence of unlabelled target domains. The obstacles in this problem are both domain shift and catastrophic forgetting. We propose Gradient Regularized Contrastive Learning (GRCL) to solve the obstacles. At the core of our method, gradient regularization plays two key roles: (1) enforcing the gradient not to harm the discriminative ability of source features which can, in turn, benefit the adaptation ability of the model to target domains; (2) constraining the gradient not to increase the classification loss on old target domains, which enables the model to preserve the performance on old target domains when adapting to an in-coming target domain. Experiments on Digits, DomainNet and Office-Caltech benchmarks demonstrate the strong performance of our approach when compared to the state-of-the-art.

Most weakly supervised fine-grained image recognition (WFGIR) approaches predominantly focus on learning the discriminative details which contain the visual variances and position clues. The position clues can be indirectly learnt by utilizing context information of discriminative visual content. However, this will cause the selected discriminative regions containing some non-discriminative information introduced by the position clues. These analysis motivate us to directly introduce position clues into visual content to only focus on the visual variances, achieving more precise discriminative region localization. Though important, position modelling usually requires significant pixel/region annotations and therefore is labor-intensive. To address this issue, we propose an end-to-end Dynamic Position-aware Network (DP-Net) to directly incorporate the position clues into visual content and dynamically align them without extra annotations, which eliminates the effect of position information for visual variances of subcategories. In particular, the DP-Net consists of: 1) Position Encoding Module, which learns a set of position-aware parts by directly adding the learnable position information into the horizontal/vertical visual content of images; 2) Position-vision Aligning Module, which dynamically aligns both visual content and learnable position information via performing graph convolution on position-aware parts; 3) Position-vision Reorganization Module, which projects the aligned position clues and visual content into the Euclidean space to construct a position-aware feature maps. Finally, the position-aware feature maps are used which is implicitly applied the aligned visual content and position clues for more accurate discriminative regions localization. Extensive experiments verify that DP-Net yields the best performance under the same settings with most competitive approaches, on CUB Bird, Stanford-Cars, and FGVC Aircraft datasets.

As a variant of standard convolution, a dilated convolution can control effective receptive fields and handle large scale variance of objects without introducing additional computational costs. To fully explore the potential of dilated convolution, we proposed a new type of dilated convolution (referred to as inception convolution), where the convolution operations have independent dilation patterns among different axes, channels and layers. To develop a practical method for learning complex inception convolution based on the data, a simple but effective search algorithm, referred to as efficient dilation optimization (EDO), is developed. Based on statistical optimization, the EDO method operates in a low-cost manner and is extremely fast when it is applied on large scale datasets. Empirical results validate that our method achieves consistent performance gains for image recognition, object detection, instance segmentation, human detection, and human pose estimation. For instance, by simply replacing the 3 × 3 standard convolution in the ResNet-50 backbone with inception convolution, we significantly improve the AP of Faster R-CNN from 36.4% to 39.2% on MS COCO.

Estimating 3D bounding boxes from monocular images is an essential component in autonomous driving, while accurate 3D object detection from this kind of data is very challenging. In this work, by intensive diagnosis experiments, we quantify the impact introduced by each sub-task and found the ‘localization error’ is the vital factor in restricting monocular 3D detection. Besides, we also investigate the underlying reasons behind localization errors, analyze the issues they might bring, and propose three strategies. First, we revisit the misalignment between the center of the 2D bounding box and the projected center of the 3D object, which is a vital factor leading to low localization accuracy. Second, we observe that accurately localizing distant objects with existing technologies is almost impossible, while those samples will mislead the learned network. To this end, we propose to remove such samples from the training set for improving the overall performance of the detector. Lastly, we also propose a novel 3D IoU oriented loss for the size estimation of the object, which is not affected by ‘localization error’. We conduct extensive experiments on the KITTI dataset, where the proposed method achieves real-time detection and outperforms previous methods by a large margin. The code will be made available at: https://github.com/xinzhuma/monodle.

Graph-neural-networks (GNN) is a rising trend for fewshot learning. A critical component in GNN is the affinity. Typically, affinity in GNN is mainly computed in the feature space, e.g., pairwise features, and does not take fully advantage of semantic labels associated to these features. In this paper, we propose a novel Mutual CRF-GNN (MCGN). In this MCGN, the labels and features of support data are used by the CRF for inferring GNN affinities in a principled and probabilistic way. Specifically, we construct a Conditional Random Field (CRF) conditioned on labels and features of support data to infer a affinity in the label space. Such affinity is fed to the GNN as the node-wise affinity. GNN and CRF mutually contributes to each other in MCGN. For GNN, CRF provides valuable affinity information. For CRF, GNN provides better features for inferring affinity. Experimental results show that our approach outperforms stateof-the-arts on datasets miniImageNet, tieredImageNet, and CIFAR-FS on both 5-way 1-shot and 5-way 5-shot settings.

Deep neural networks achieve state-of-the-art and sometimes super-human performance across various domains. However, when learning tasks sequentially, the networks easily forget the knowledge of previous tasks, known as "catastrophic forgetting". To achieve the consistencies between the old tasks and the new task, one effective solution is to modify the gradient for update. Previous methods enforce independent gradient constraints for different tasks, while we consider these gradients contain complex information, and propose to leverage inter-task information by gradient decomposition. In particular, the gradient of an old task is decomposed into a part shared by all old tasks and a part specific to that task. The gradient for update should be close to the gradient of the new task, consistent with the gradients shared by all old tasks, and orthogonal to the space spanned by the gradients specific to the old tasks. In this way, our approach encourages common knowledge consolidation without impairing the task-specific knowledge. Furthermore, the optimization is performed for the gradients of each layer separately rather than the concatenation of all gradients as in previous works. This effectively avoids the influence of the magnitude variation of the gradients in different layers. Extensive experiments validate the effectiveness of both gradient-decomposed optimization and layer-wise updates. Our proposed method achieves state-of-the-art results on various benchmarks of continual learning.

Human pose estimation has achieved significant progress in recent years. However, most of the recent methods focus on improving accuracy using complicated models and ignoring real-time efficiency. To achieve a better trade-off between accuracy and efficiency, we propose a novel neural architecture search (NAS) method, termed ViP-NAS, to search networks in both spatial and temporal levels for fast online video pose estimation. In the spatial level, we carefully design the search space with five different dimensions including network depth, width, kernel size, group number, and attentions. In the temporal level, we search from a series of temporal feature fusions to optimize the total accuracy and speed across multiple video frames. To the best of our knowledge, we are the first to search for the temporal feature fusion and automatic computation allocation in videos. Extensive experiments demonstrate the effectiveness of our approach on the challenging COCO2017 and PoseTrack2018 datasets. Our discovered model family, S-ViPNAS and T-ViPNAS, achieve significantly higher inference speed (CPU real-time) without sacrificing the accuracy compared to the previous state-of-the-art methods.

We present BN-NAS, neural architecture search with Batch Normalization (BN-NAS), to accelerate neural architecture search (NAS). BN-NAS can significantly reduce the time required by model training and evaluation in NAS. Specifically, for fast evaluation, we propose a BN-based indicator for predicting subnet performance at a very early training stage. The BN-based indicator further facilitates us to improve the training efficiency by only training the BN parameters during the supernet training. This is based on our observation that training the whole supernet is not necessary while training only BN parameters accelerates network convergence for network architecture search. Extensive experiments show that our method can significantly shorten the time of training supernet by more than 10 times and shorten the time of evaluating subnets by more than 600,000 times without losing accuracy. The source codes are available at https://github.com/bychen515/BNNAS.

We introduce the first Neural Architecture Search (NAS) method to find a better transformer architecture for image recognition. Recently, transformers without CNN-based backbones are found to achieve impressive performance for image recognition. However, the transformer is designed for NLP tasks and thus could be sub-optimal when directly used for image recognition. In order to improve the visual representation ability for transformers, we propose a new search space and searching algorithm. Specifically, we introduce a locality module that models the local correlations in images explicitly with fewer computational cost. With the locality module, our search space is defined to let the search algorithm freely trade off between global and local information as well as optimizing the low-level design choice in each module. To tackle the problem caused by huge search space, a hierarchical neural architecture search method is proposed to search the optimal vision transformer from two levels separately with the evolutionary algorithm. Extensive experiments on the ImageNet dataset demonstrate that our method can find more discriminative and efficient trans-former variants than the ResNet family (e.g., ResNet101) and the baseline ViT for image classification. The source codes are available at https://github.com/bychen515/GLiT.

Automation of neural architecture design has been a coveted alternative to human experts. Various search methods have been proposed aiming to find the optimal architecture in the search space. One would expect the search results to improve when the search space grows larger since it would potentially contain more performant candidates. Surprisingly, we observe that enlarging search space is unbeneficial or even detrimental to existing NAS methods such as DARTS, ProxylessNAS, and SPOS. This counterintuitive phenomenon suggests that enabling existing methods to large search space regimes is non-trivial. However, this problem is less discussed in the literature.We present a Neural Search-space Evolution (NSE) scheme, the first neural architecture search scheme designed especially for large space neural architecture search problems. The necessity of a well-designed search space with constrained size is a tacit consent in existing methods, and our NSE aims at minimizing such necessity. Specifically, the NSE starts with a search space subset, then evolves the search space by repeating two steps: 1) search an optimized space from the search space subset, 2) refill this subset from a large pool of operations that are not traversed. We further extend the flexibility of obtainable architectures by introducing a learnable multi-branch setting. With the proposed method, we achieve 77.3% top-1 retrain accuracy on ImageNet with 333M FLOPs, which yielded a state-of-the-art performance among previous auto-generated architectures that do not involve knowledge distillation or weight pruning. When the latency constraint is adopted, our result also performs better than the previous best-performing mobile models with a 77.9% Top-1 retrain accuracy. Code is available at https://github.com/orashi/NSENAS.

Geometry Projection is a powerful depth estimation method in monocular 3D object detection. It estimates depth dependent on heights, which introduces mathematical priors into the deep model. But projection process also introduces the error amplification problem, in which the error of the estimated height will be amplified and reflected greatly at the output depth. This property leads to uncontrollable depth inferences and also damages the training efficiency. In this paper, we propose a Geometry Uncertainty Projection Network (GUP Net) to tackle the error amplification problem at both inference and training stages. Specifically, a GUP module is proposed to obtains the geometry-guided uncertainty of the inferred depth, which not only provides high reliable confidence for each depth but also benefits depth learning. Furthermore, at the training stage, we propose a Hierarchical Task Learning strategy to reduce the instability caused by error amplification. This learning algorithm monitors the learning situation of each task by a proposed indicator and adaptively assigns the proper loss weights for different tasks according to their pre-tasks situation. Based on that, each task starts learning only when its pre-tasks are learned well, which can significantly improve the stability and efficiency of the training process. Extensive experiments demonstrate the effectiveness of the proposed method. The overall model can infer more reliable object depth than existing methods and outperforms the state-of-the-art image-based monocular 3D detectors by 3.74% and 4.7% AP40of the car and pedestrian categories on the KITTI benchmark. The code and model will be released at https://github.com/SuperMHP/GUPNet.

Quantization Neural Networks (QNN) have attracted a lot of attention due to their high efficiency. To enhance the quantization accuracy, prior works mainly focus on designing advanced quantization algorithms but still fail to achieve satisfactory results under the extremely low-bit case. In this work, we take an architecture perspective to investigate the potential of high-performance QNN. Therefore, we propose to combine Network Architecture Search methods with quantization to enjoy the merits of the two sides. However, a naive combination inevitably faces unacceptable time consumption or unstable training problem. To alleviate these problems, we first propose the joint training of architecture and quantization with a shared step size to acquire a large number of quantized models. Then a bit-inheritance scheme is introduced to transfer the quantized models to the lower bit, which further reduces the time cost and meanwhile improves the quantization accuracy. Equipped with this overall framework, dubbed as Once Quantization-Aware Training (OQAT), our searched model family, OQATNets, achieves a new state-of-the-art compared with various architectures under different bit-widths. In particular, OQAT-2bit-M achieves 61.6% ImageNet Top-1 accuracy, outperforming 2-bit counterpart MobileNetV3 by a large margin of 9% with 10% less computation cost. A series of quantization-friendly architectures are identified easily and extensive analysis can be made to summarize the interaction between quantization and neural architectures. Codes and models are released at https://github.com/LaVieEnRoseSMZ/OQA

Aggregating features from different depths of a network is widely adopted to improve the network capability. Lots of modern architectures are equipped with skip connections, which actually makes the feature aggregation happen in all these networks. Since different features tell different semantic meanings, there are inconsistencies and incompatibilities to be solved. However, existing works naively blend deep features via element-wise summation or concatenation with a convolution behind. Better feature aggregation method beyond summation or concatenation is rarely explored. In this paper, given two layers of features to be aggregated together, we first detect and identify where and what needs to be updated in one layer, then replace the feature at the identified location with the information of the other layer This process, which we call DEtect-rePLAce (DEPLA), enables us to avoid inconsistent patterns while keeping useful information in the merged outputs. Experimental results demonstrate our method largely boosts multiple baselines e.g. ResNet, FishNet and FPN on three major vision tasks including ImageNet classification, MS COCO object detection and instance segmentation.

This paper studies the task of estimating the 3D human poses of multiple persons from multiple calibrated camera views. Following the top-down paradigm, we decompose the task into two stages, i.e. person localization and pose estimation. Both stages are processed in coarse-to-fine manners. And we propose three task-specific graph neural networks for effective message passing. For 3D person localization, we first use Multi-view Matching Graph Module (MMG) to learn the cross-view association and recover coarse human proposals. The Center Refinement Graph Module (CRG) further refines the results via flexible point-based prediction. For 3D pose estimation, the Pose Regression Graph Module (PRG) learns both the multi-view geometry and structural relations between human joints. Our approach achieves state-of-the-art performance on CMU Panoptic and Shelf datasets with significantly lower computation complexity.

Semantic segmentation is a challenging task in the absence of densely labelled data. Only relying on class activation maps (CAM) with image-level labels provides deficient segmentation supervision. Prior works thus consider pre-trained models to produce coarse saliency maps to guide the generation of pseudo segmentation labels. However, the commonly used off-line heuristic generation process cannot fully exploit the benefits of these coarse saliency maps. Motivated by the significant inter-task correlation, we propose a novel weakly supervised multi-task framework termed as AuxSegNet, to leverage saliency detection and multi-label image classification as auxiliary tasks to improve the primary task of semantic segmentation using only image-level ground-truth labels. Inspired by their similar structured semantics, we also propose to learn a cross-task global pixellevel affinity map from the saliency and segmentation representations. The learned cross-task affinity can be used to refine saliency predictions and propagate CAM maps to provide improved pseudo labels for both tasks. The mutual boost between pseudo label updating and cross-task affinity learning enables iterative improvements on segmentation performance. Extensive experiments demonstrate the effectiveness of the proposed auxiliary learning network structure and the cross-task affinity learning method. The proposed approach achieves state-of-the-art weakly supervised segmentation performance on the challenging PASCAL VOC 2012 and MS COCO benchmarks.1

Regression-based methods have recently shown promising results in reconstructing human meshes from monocular images. By directly mapping raw pixels to model parameters, these methods can produce parametric models in a feed-forward manner via neural networks. However, minor deviation in parameters may lead to noticeable mis-alignment between the estimated meshes and image evidences. To address this issue, we propose a Pyramidal Mesh Alignment Feedback (PyMAF) loop to leverage a feature pyramid and rectify the predicted parameters explicitly based on the mesh-image alignment status in our deep regressor. In PyMAF, given the currently predicted parameters, mesh-aligned evidences will be extracted from finer-resolution features accordingly and fed back for parameter rectification. To reduce noise and enhance the reliability of these evidences, an auxiliary pixel-wise supervision is imposed on the feature encoder, which provides mesh-image correspondence guidance for our network to preserve the most related information in spatial features. The efficacy of our approach is validated on several benchmarks, including Human3.6M, 3DPW, LSP, and COCO, where experimental results show that our approach consistently improves the mesh-image alignment of the reconstruction. The project page with code and video results can be found at https://hongwenzhang.github.io/pymaf.

Data sampling acts as a pivotal role in training deep learning models. However, an effective sampling schedule is difficult to learn due to its inherent high-dimension as a hyper-parameter. In this paper, we propose an AutoSampling method to automatically learn sampling schedules for model training, which consists of the multi-exploitation step aiming for optimal local sampling schedules and the exploration step for the ideal sampling distribution. More specifically, we achieve sampling schedule search with shortened exploitation cycle to provide enough supervision. In addition, we periodically estimate the sampling distribution from the learned sampling schedules and perturb it to search in the distribution space. The combination of two searches allows us to learn a robust sampling schedule. We apply our AutoSampling method to a variety of image classification tasks illustrating the effectiveness of the proposed method.

Automated data augmentation (ADA) techniques have played an important role in boosting the performance of deep models. Such techniques mostly aim to optimize a parameterized distribution over a discrete augmentation space. Thus, are restricted by the discretization of the search space which normally is handcrafted. To overcome the limitations, we take the first step to constructing a continuous mapping from $\mathbb{R}^d$ to image transformations (an augmentation space). Using this mapping, we take a novel approach where 1) we pose the ADA as a continuous optimization problem over the parameters of the augmentation distribution; and 2) use Stochastic Gradient Langevin Dynamics to learn and sample augmentations. This allows us to potentially explore the space of infinitely many possible augmentations, which otherwise was not possible due to the discretization of the space. This view of ADA is radically different from the standard discretization based view of ADA, and it opens avenues for utilizing the vast efficient gradient-based algorithms available for continuous optimization problems. Results over multiple benchmarks demonstrate the efficiency improvement of this work compared with previous methods.

Traditional video compression approaches build upon the hybrid coding framework with motion-compensated prediction and residual transform coding. In this paper, we propose the first end-to-end deep video compression framework to take advantage of both the classical compression architecture and the powerful non-linear representation ability of neural networks. Our framework employs pixel-wise motion information, which is learned from an optical flow network and further compressed by an auto-encoder network to save bits. The other compression components are also implemented by the well-designed networks for high efficiency. All the modules are jointly optimized by using the rate-distortion trade-off and can collaborate with each other. More importantly, the proposed deep video compression framework is very flexible and can be easily extended by using lightweight or advanced networks for higher speed or better efficiency. We also propose to introduce the adaptive quantization layer to reduce the number of parameters for variable bitrate coding. Comprehensive experimental results demonstrate the effectiveness of the proposed framework on the benchmark datasets.

Spatio-temporal action localization consists of three levels of tasks: spatial localization, action classification, and temporal localization. In this work, we propose a new progressive cross-stream cooperation (PCSC) framework that improves all three tasks above. The basic idea is to utilize both spatial region (resp., temporal segment proposals) and features from one stream (i.e., the Flow/RGB stream) to help another stream (i.e., the RGB/Flow stream) to iteratively generate better bounding boxes in the spatial domain (resp., temporal segments in the temporal domain). In this way, not only the actions could be more accurately localized both spatially and temporally, but also the action classes could be predicted more precisely. Specifically, we first combine the latest region proposals (for spatial detection) or segment proposals (for temporal localization) from both streams to form a larger set of labelled training samples to help learn better action detection or segment detection models. Second, to learn better representations, we also propose a new message passing approach to pass information from one stream to another stream, which also leads to better action detection and segment detection models. By first using our newly proposed PCSC framework for spatial localization at the frame-level and then applying our temporal PCSC framework for temporal localization at the tube-level, the action localization results are progressively improved at both the frame level and the video level. Comprehensive experiments on two benchmark datasets UCF-101-24 and J-HMDB demonstrate the effectiveness of our newly proposed approaches for spatio-temporal action localization in realistic scenarios.

This paper proposes a new unsupervised domain adaptation approach called Collaborative and Adversarial Network (CAN), which uses the domain-collaborative and domain-adversarial learning strategies for training the neural network. The domain-collaborative learning strategy aims to learn domain specific feature representation to preserve the discriminability for the target domain, while the domain adversarial learning strategy aims to learn domain invariant feature representation to reduce the domain distribution mismatch between the source and target domains. We show that these two learning strategies can be uniformly formulated as domain classifier learning with positive or negative weights on the losses. We then design a collaborative and adversarial training scheme, which automatically learns domain specific representations from lower blocks in CNNs through collaborative learning and domain invariant representations from higher blocks through adversarial learning. Moreover, to further enhance the discriminability in the target domain, we propose Self-Paced CAN (SPCAN), which progressively selects pseudo-labeled target samples for re-training the classifiers. We employ a self-paced learning strategy such that we can select pseudo-labeled target samples in an easy-to-hard fashion. Additionally, we build upon the popular two-stream approach to extend our domain adaptation approach for more challenging video action recognition task, which additionally considers the cooperation between the RGB stream and the optical flow stream. We propose the Two-stream SPCAN (TS-SPCAN) method to select and reweight the pseudo labeled target samples of one stream (RGB/Flow) based on the information from the other stream (Flow/RGB) in a cooperative way. As a result, our TS-SPCAN model is able to exchange the information between the two streams. Comprehensive experiments on different benchmark datasets, Office-31, ImageCLEF-DA and VISDA-2017 for the object recognition task, and UCF101-10 and HMDB51-10 for the video action recognition task, show our newly proposed approaches achieve the state-of-the-art performance, which clearly demonstrates the effectiveness of our proposed approaches for unsupervised domain adaptation.

In this work, we propose a simple but effective channel pruning framework called Progressive Channel Pruning (PCP) to accelerate Convolutional Neural Networks (CNNs). In contrast to the existing channel pruning methods that prune the channels only once per layer in a layer-by-layer fashion, our new progressive framework iteratively prunes a small number of channels from several selected layers, which consists of a three-step attempting-selecting-pruning pipeline in each iteration. In the attempting step, we attempt to prune a pre-defined number of channels from one layer by using any existing channel pruning methods and estimate the accuracy drop for this layer based on the labelled samples in the validation set. In the selecting step, based on the estimated accuracy drops for all layers, we propose a greedy strategy to automatically select a set of layers that will lead to less overall accuracy drop after pruning these layers. In the pruning step, we prune a small number of channels from these selected layers. We further extend our PCP framework to prune channels for the deep transfer learning methods like Domain Adversarial Neural Network (DANN), in which we effectively reduce the data distribution mismatch in the channel pruning process by using both labelled samples from the source domain and pseudo-labelled samples from the target domain. Our comprehensive experiments on two benchmark datasets demonstrate that our PCP framework outperforms the existing channel pruning approaches under both supervised learning and transfer learning settings.

As a challenging task of high-level video understanding, weakly supervised temporal action localization has attracted more attention recently. Due to the usage of video-level category labels, this task is usually formulated as the task of classification, which always suffers from the contradiction between classification and detection. In this paper, we describe a novel approach to alleviate the contradiction for detecting more complete action instances by explicitly modeling sub-actions. Our method makes use of three innovations to model the latent sub-actions. First, our framework uses prototypes to represent sub-actions, which can be automatically learned in an end-to-end way. Second, we regard the relations among sub-actions as a graph, and construct the correspondences between sub-actions and actions by the graph pooling operation. Doing so not only makes the sub-actions inter-dependent to facilitate the multi-label setting, but also naturally use the video-level labels as weak supervision. Third, we devise three complementary loss functions, namely, representation loss, balance loss and relation loss to ensure the learned sub-actions are diverse and have clear semantic meanings. Experimental results on THUMOS14 and ActivityNet1.3 datasets demonstrate the effectiveness of our method and superior performance over state-of-the-art approaches.

Extreme instance imbalance among categories and combinatorial explosion make the recognition of Human-Object Interaction (HOI) a challenging task. Few studies have addressed both challenges directly. Motivated by the success of few-shot learning that learns a robust model from a few instances, we formulate HOI as a few-shot task in a meta-learning framework to alleviate the above challenges. Due to the fact that the intrinsical characteristic of HOI is diverse and interactive, we propose a Semantic-guided Attentive Prototypes Network (SAPNet) framework to learn a semantic-guided metric space where HOI recognition can be performed by computing distances to attentive prototypes of each class. Specifically, the model generates attentive prototypes guided by the category names of actions and objects, which highlight the commonalities of images from the same class in HOI. In addition, we design two alternative prototypes calculation methods, i.e., Prototypes Shift (PS) approach and Hallucinatory Graph Prototypes (HGP) approach, which explore to learn a suitable category prototypes representations in HOI. Finally, in order to realize the task of few-shot HOI, we reorganize 2 HOI benchmark datasets with 2 split strategies, i.e., HICO-NN, TUHOI-NN, HICO-NF, and TUHOI-NF. Extensive experimental results on these datasets have demonstrated the effectiveness of our proposed SAPNet approach.

The existing neural architecture search (NAS) methods usually restrict the search space to the pre-defined types of block for a fixed macro-architecture. However, this strategy will limit the search space and affect architecture flexibility if block proposal search (BPS) is not considered for NAS. As a result, block structure search is the bottleneck in many previous NAS works. In this work, we propose a new evolutionary algorithm referred to as latency EvoNAS (LEvoNAS) for block structure search, and also incorporate it to the NAS framework by developing a novel two-stage framework referred to as Block Proposal NAS (BP-NAS). Comprehensive experimental results on two computer vision tasks demonstrate the superiority of our newly proposed approach over the state-of-the-art lightweight methods. For the classification task on the ImageNet dataset, our BPN-A is better than 1.0-MobileNetV2 with similar latency, and our BPN-B saves 23.7% latency when compared with 1.4-MobileNetV2 with higher top-1 accuracy. Furthermore, for the object detection task on the COCO dataset, our method achieves significant performance improvement than MobileNetV2, which demonstrates the generalization capability of our newly proposed framework.

There are two major lines of works, i.e., anchor-based and frame-based approaches, in the field of temporal action localization. But each line of works is inherently limited to a certain detection granularity and cannot simultaneously achieve high recall rates with accurate action boundaries. In this work, we propose a progressive cross-granularity cooperation (PCG-TAL) framework to effectively take advantage of complementarity between the anchor-based and frame-based paradigms, as well as between two-view clues (i.e., appearance and motion). Specifically, our new Anchor-Frame Cooperation (AFC) module can effectively integrate both two-granularity and two-stream knowledge at the feature and proposal levels, as well as within each AFC module and across adjacent AFC modules. Specifically, the RGB-stream AFC module and the flow-stream AFC module are stacked sequentially to form a progressive localization framework. The whole framework can be learned in an end-to-end fashion, whilst the temporal action localization performance can be gradually boosted in a progressive manner. Our newly proposed framework outperforms the state-of-the-art methods on three benchmark datasets the THUMOS14, ActivityNet v1.3 and UCF-101-24, which clearly demonstrates the effectiveness of our framework.

Weakly supervised temporal action localization is a challenging task as only the video-level annotation is available during the training process. To address this problem, we propose a two-stage approach to generate high-quality frame-level pseudo labels by fully exploiting multi-resolution information in the temporal domain and complementary information between the appearance (i.e., RGB) and motion (i.e., optical flow) streams. In the first stage, we propose an Initial Label Generation (ILG) module to generate reliable initial frame-level pseudo labels. Specifically, in this newly proposed module, we exploit temporal multi-resolution consistency and cross-stream consistency to generate high quality class activation sequences (CASs), which consist of a number of sequences with each sequence measuring how likely each video frame belongs to one specific action class. In the second stage, we propose a Progressive Temporal Label Refinement (PTLR) framework to iteratively refine the pseudo labels, in which we use a set of selected frames with highly confident pseudo labels to progressively train two networks and better predict action class scores at each frame. Specifically, in our newly proposed PTLR framework, two networks called Network-OTS and Network-RTS, which are respectively used to generate CASs for the original temporal scale and the reduced temporal scales, are used as two streams (i.e., the OTS stream and the RTS stream) to refine the pseudo labels in turn. By this way, multi-resolution information in the temporal domain is exchanged at the pseudo label level, and our work can help improve each network/stream by exploiting the refined pseudo labels from another network/stream. Comprehensive experiments on two benchmark datasets THUMOS14 and ActivityNet v1.3 demonstrate the effectiveness of our newly proposed method for weakly supervised temporal action localization.

Pedestrian detection is a challenging and hot research topic in the field of computer vision, especially for the crowded scenes where occlusion happens frequently. In this paper, we propose a novel AutoPedestrian scheme that automatically augments the pedestrian data and searches for suitable loss functions, aiming for better performance of pedestrian detection especially in crowded scenes. To our best knowledge, it is the first work to automatically search the optimal policy of data augmentation and loss function jointly for the pedestrian detection. To achieve the goal of searching the optimal augmentation scheme and loss function jointly, we first formulate the data augmentation policy and loss function as probability distributions based on different hyper-parameters. Then, we apply a double-loop scheme with importance-sampling to solve the optimization problem of data augmentation and loss function types efficiently. Comprehensive experiments on two popular benchmarks of CrowdHuman and CityPersons show the effectiveness of our proposed method. In particular, we achieve 40.58% in MR on CrowdHuman datasets and 11.3% in MR on CityPersons reasonable subset, yielding new state-of-the-art results on these two datasets.

In this work, we propose a new generic multi-modality domain adaptation framework called Progressive Modality Cooperation (PMC) to transfer the knowledge learned from the source domain to the target domain by exploiting multiple modality clues (e.g., RGB and depth) under the multi-modality domain adaptation (MMDA) and the more general multi-modality domain adaptation using privileged information (MMDA-PI) settings. Under the MMDA setting, the samples in both domains have all the modalities. Through effective collaboration among multiple modalities, the two newly proposed modules in our PMC can select the reliable pseudo-labeled target samples, which captures the modality-specific information and modality-integrated information, respectively. Under the MMDA-PI setting, some modalities are missing in the target domain. Hence, to better exploit the multi-modality data in the source domain, we further propose the PMC with privileged information (PMC-PI) method by proposing a new multi-modality data generation (MMG) network. MMG generates the missing modalities in the target domain based on the source domain data by considering both domain distribution mismatch and semantics preservation, which are respectively achieved by using adversarial learning and conditioning on weighted pseudo semantic class labels. Extensive experiments on three image datasets and eight video datasets for various multi-modality cross-domain visual recognition tasks under both MMDA and MMDA-PI settings clearly demonstrate the effectiveness of our proposed PMC framework.

Recently, dense video captioning has made attractive progress in detecting and captioning all events in a long untrimmed video. Despite promising results were achieved, most existing methods do not sufficiently explore the scene evolution within an event temporal proposal for captioning, and therefore perform less satisfactorily when the scenes and objects change over a relatively long proposal. To address this problem, we propose a graph-based partition-and-summarization (GPaS) framework for dense video captioning within two stages. For the “partition” stage, a whole event proposal is split into short video segments for captioning at a finer level. For the “summarization” stage, the generated sentences carrying rich description information for each segment are summarized into one sentence to describe the whole event. We particularly focus on the “summarization” stage, and propose a framework that effectively exploits the relationship between semantic words for summarization. We achieve this goal by treating semantic words as the nodes in a graph and learning their interactions by coupling Graph Convolutional Network (GCN) and Long Short Term Memory (LSTM), with the aid of visual cues. Two schemes of GCN-LSTM Interaction (GLI) modules are proposed for seamless integration of GCN and LSTM. The effectiveness of our approach is demonstrated via an extensive comparison with the state-of-the-arts methods on the two benchmarks ActivityNet Captions dataset and YouCook II dataset.

Person Search is a challenging task which requires to retrieve a person's image and the corresponding position from an image dataset. It consists of two sub-tasks: pedestrian detection and person re-identification (re-ID). One of the key challenges is to properly combine the two sub-tasks into a unified framework. Existing works usually adopt a straightforward strategy by concatenating a detector and a re-ID model directly, either into an integrated model or into separated models. We argue that simply concatenating detection and re-ID is a sub-optimal solution, and we propose a Hierarchical Online Instance Matching (HOIM) loss which exploits the hierarchical relationship between detection and re-ID to guide the learning of our network. Our novel HOIM loss function harmonizes the objectives of the two sub-tasks and encourages better feature learning. In addition, we improve the loss update policy by introducing Selective Memory Refreshment (SMR) for unlabeled persons, which takes advantage of the potential discrimination power of unlabeled data. From the experiments on two standard person search benchmarks, i.e. CUHK-SYSU and PRW, we achieve state-of-the-art performance, which justifies the effectiveness of our proposed HOIM loss on learning robust features.

In this paper, we propose an online multi-object tracking (MOT) approach that integrates data association and single object tracking (SOT) with a unified convolutional network (ConvNet), named DASOTNet. The intuition behind integrating data association and SOT is that they can complement each other. Following Siamese network architecture, DASOTNet consists of the shared feature ConvNet, the data association branch and the SOT branch. Data association is treated as a special re-identification task and solved by learning discriminative features for different targets in the data association branch. To handle the problem that the computational cost of SOT grows intolerably as the number of tracked objects increases, we propose an efficient two-stage tracking method in the SOT branch, which utilizes the merits of correlation features and can simultaneously track all the existing targets within one forward propagation. With feature sharing and the interaction between them, data association branch and the SOT branch learn to better complement each other. Using a multi-task objective, the whole network can be trained end-to-end. Compared with state-of-the-art online MOT methods, our method is much faster while maintaining a comparable performance.

In this work, we propose a new layer-by-layer channel pruning method called Channel Pruning guided by classification Loss and feature Importance (CPLI). In contrast to the existing layer-by-layer channel pruning approaches that only consider how to reconstruct the features from the next layer, our approach additionally take the classification loss into account in the channel pruning process. We also observe that some reconstructed features will be removed at the next pruning stage. So it is unnecessary to reconstruct these features. To this end, we propose a new strategy to suppress the influence of unimportant features (i.e., the features will be removed at the next pruning stage). Our comprehensive experiments on three benchmark datasets, i.e., CIFAR-10, ImageNet, and UCF-101, demonstrate the effectiveness of our CPLI method.

Recently, graph convolutional networks have achieved remarkable performance for skeleton-based action recognition. In this work, we identify a problem posed by the GCNs for skeleton-based action recognition, namely part-level action modeling. To address this problem, a novel Part-Level Graph Convolutional Network (PL-GCN) is proposed to capture part-level information of skeletons. Different from previous methods, the partition of body parts is learnable rather than manually defined. We propose two part-level blocks, namely Part Relation block (PR block) and Part Attention block (PA block), which are achieved by two differentiable operations, namely graph pooling operation and graph unpooling operation. The PR block aims at learning high-level relations between body parts while the PA block aims at highlighting the important body parts in the action. Integrating the original GCN with the two blocks, the PL-GCN can learn both part-level and joint-level information of the action. Extensive experiments on two benchmark datasets show the state-of-the-art performance on skeleton-based action recognition and demonstrate the effectiveness of the proposed method.

In this paper, we propose a weakly supervised temporal action localization method on untrimmed videos based on prototypical networks. We observe two challenges posed by weakly supervision, namely action-background separation and action relation construction. Unlike the previous method, we propose to achieve action-background separation only by the original videos. To achieve this, a clustering loss is adopted to separate actions from backgrounds and learn intra-compact features, which helps in detecting complete action instances. Besides, a similarity weighting module is devised to further separate actions from backgrounds. To effectively identify actions, we propose to construct relations among actions for prototype learning. A GCN-based prototype embedding module is introduced to generate relational prototypes. Experiments on THUMOS14 and ActivityNet1.2 datasets show that our method outperforms the state-of-the-art methods.

In this work, we propose a unified model compression framework called Multi-Dimensional Pruning (MDP) to simultaneously compress the convolutional neural networks (CNNs) on multiple dimensions. In contrast to the existing model compression methods that only aim to reduce the redundancy along either the spatial/spatial-temporal dimension (e.g., spatial dimension for 2D CNNs, spatial and temporal dimensions for 3D CNNs) or the channel dimension, our newly proposed approach can simultaneously reduce the spatial/spatial-temporal and the channel redundancies for CNNs. Specifically, in order to reduce the redundancy along the spatial/spatial-temporal dimension, we downsample the input tensor of a convolutional layer, in which the scaling factor for the downsampling operation is adaptively selected by our approach. After the convolution operation, the output tensor is upsampled to the original size to ensure the unchanged input size for the subsequent CNN layers. To reduce the channel-wise redundancy, we introduce a gate for each channel of the output tensor as its importance score, in which the gate value is automatically learned. The channels with small importance scores will be removed after the model compression process. Our comprehensive experiments on four benchmark datasets demonstrate that our MDP framework outperforms the existing methods when pruning both 2D CNNs and 3D CNNs.

Neural architecture search (NAS) has demonstrated much success in automatically designing effective neural network architectures. To improve the efficiency of NAS, previous approaches adopt weight sharing method to force all models share the same set of weights. However, it has been observed that a model performing better with shared weights does not necessarily perform better when trained alone. In this paper, we analyse existing weight sharing one-shot NAS approaches from a Bayesian point of view and identify the Posterior Fading problem, which compromises the effectiveness of shared weights. To alleviate this problem, we present a novel approach to guide the parameter posterior towards its true distribution. Moreover, a hard latency constraint is introduced during the search so that the desired latency can be achieved. The resulted method, namely Posterior Convergent NAS (PC-NAS), achieves state-of-the-art performance under standard GPU latency constraint on ImageNet.

Spatial-temporal graphs have been widely used by skeleton-based action recognition algorithms to model human action dynamics. To capture robust movement patterns from these graphs, long-range and multi-scale context aggregation and spatial-temporal dependency modeling are critical aspects of a powerful feature extractor. However, existing methods have limitations in achieving (1) unbiased long-range joint relationship modeling under multi-scale operators and (2) unobstructed cross-spacetime information flow for capturing complex spatial-temporal dependencies. In this work, we present (1) a simple method to disentangle multi-scale graph convolutions and (2) a unified spatial-temporal graph convolutional operator named G3D. The proposed multi-scale aggregation scheme disentangles the importance of nodes in different neighborhoods for effective long-range modeling. The proposed G3D module leverages dense cross-spacetime edges as skip connections for direct information propagation across the spatial-temporal graph. By coupling these proposals, we develop a powerful feature extractor named MS-G3D based on which our model outperforms previous state-of-the-art methods on three large-scale datasets: NTU RGB+D 60, NTU RGB+D 120, and Kinetics Skeleton 400.

Object recognition techniques using convolutional neural networks (CNN) have achieved great success. However, state-of-the-art object detection methods still perform poorly on large vocabulary and long-tailed datasets, e.g. LVIS. In this work, we analyze this problem from a novel perspective: each positive sample of one category can be seen as a negative sample for other categories, making the tail categories receive more discouraging gradients. Based on it, we propose a simple but effective loss, named equalization loss, to tackle the problem of long-tailed rare categories by simply ignoring those gradients for rare categories. The equalization loss protects the learning of rare categories from being at a disadvantage during the network parameter updating. Thus the model is capable of learning better discriminative features for objects of rare classes. Without any bells and whistles, our method achieves AP gains of 4.1% and 4.8% for the rare and common categories on the challenging LVIS benchmark, compared to the Mask R-CNN baseline. With the utilization of the effective equalization loss, we finally won the 1st place in the LVIS Challenge 2019. Code has been made available at: https://github.com/tztztztztz/eql.detectron2.

Estimating 3D mesh of the human body from a single 2D image is an important task with many applications such as augmented reality and Human-Robot interaction. However, prior works reconstructed 3D mesh from global image feature extracted by using convolutional neural network (CNN), where the dense correspondences between the mesh surface and the image pixels are missing, leading to suboptimal solution. This paper proposes a model-free 3D human mesh estimation framework, named DecoMR, which explicitly establishes the dense correspondence between the mesh and the local image features in the UV space (i.e. a 2D space used for texture mapping of 3D mesh). DecoMR first predicts pixel-to-surface dense correspondence map (i.e., IUV image), with which we transfer local features from the image space to the UV space. Then the transferred local image features are processed in the UV space to regress a location map, which is well aligned with transferred features. Finally we reconstruct 3D human mesh from the regressed location map with a predefined mapping function. We also observe that the existing discontinuous UV map are unfriendly to the learning of network. Therefore, we propose a novel UV map that maintains most of the neighboring relations on the original mesh surface. Experiments demonstrate that our proposed local feature alignment and continuous UV map outperforms existing 3D mesh based methods on multiple public benchmarks. Code will be made available at https: //github.com/zengwang430521/DecoMR.

Neural Architecture Search (NAS) achieves significant progress in many computer vision tasks. While many methods are proposed to improve the efficiency of NAS, the search progress is still laborious because training and evaluating plausible architectures over large search space is time-consuming. Assessing network candidates under a proxy (i.e., computationally reduced setting) thus becomes inevitable. In this paper, we observe that most existing proxies exhibit different behaviors in maintaining the rank consistency among network candidates. In particular, some proxies can be more reliable - the rank of candidates does not differ much comparing their reduced setting performance and final performance. In this paper, we systematically investigate some widely adopted reduction factors and report our observations. Inspired by these observations, we present a reliable proxy and further formulate a hierarchical proxy strategy that spends more computations on candidate networks that are potentially more accurate, while discards unpromising ones in early stage with a fast proxy. This leads to an economical evolutionary-based NAS (EcoNAS), which achieves an impressive 400×search time reduction in comparison to the evolutionary-based state of the art [19] (8 v.s. 3150 GPU days). Some new proxies led by our observations can also be applied to accelerate other NAS methods while still able to discover good candidate networks with performance matching those found by previous proxy strategies. Codes and models will be released to facilitate future research.

Learning an optimal policy for autonomous driving task to confront with complex environment is a long- studied challenge. Imitative reinforcement learning is accepted as a promising approach to learn a robust driving policy through expert demonstrations and interactions with environments. However, this model utilizes non-smooth rewards, which have a negative impact on matching between navigation commands and trajectory (state-action pairs), and degrade the generalizability of an agent. Smooth rewards are crucial to discriminate actions generated from sub-optimal policy. In this paper, we propose a navigation command matching (NCM) model to address this issue. There are two key components in NCM, 1) a matching measurer produces smooth navigation rewards that measure matching between navigation commands and trajectory; 2) attention-guided agent performs actions given states where salient regions in RGB images (i.e. roadsides, lane markings and dynamic obstacles) are highlighted to amplify their influence on the final model. We obtain navigation rewards and store transitions to replay buffer after an episode, so NCM is able to discriminate actions generated from suboptimal policy. Experiments on CARLA driving benchmark show our proposed NCM outperforms previous state-of-the- art models on various tasks in terms of the percentage of successfully completed episodes. Moreover, our model improves generalizability of the agent and obtains good performance even in unseen scenarios.

Existing deep learning based weakly supervised fine-grained image recognition (WFGIR) methods usually pick out the discriminative regions from the high-level feature (HLF) maps directly. However, as HLF maps are derived based on spatial aggregation of convolution which is basically a pattern matching process that applies fixed filters, it is ineffective to model visual contents of same semantic but varying posture or perspective. We argue that this will cause the selected discriminative regions of same sub-category are not semantically corresponding and thus degrade the WFGIR performance. To address this issue, we propose an end-to-end Category-specific Semantic Coherency Network (CSC-Net) to semantically align the discriminative regions of the same subcategory. Specifically, CSC-Net consists of: 1) Local-to-Attribute Projecting Module (LPM), which automatically learns a set of latent attributes via collecting the category-specific semantic details while eliminating the varying spatial distributions from the local regions. 2) Latent Attribute Aligning (LAA), which aligns the latent attributes to specific semantic via graph convolution based on their discriminability, to achieve category-specific semantic coherency; 3) Attribute-to-Local Resuming Module (ARM), which resumes the original Euclidean space of latent attributes and construct latent attribute aligned feature maps by a location-embedding graph unpooling operation. Finally, the new feature maps are used which applies the category-specific semantic coherency implicitly for more accurate discriminative regions localization. Extensive experiments verify that CSC-Net yields the best performance under the same settings with most competitive approaches, on CUB Bird, Stanford-Cars, and FGVC Aircraft datasets.

The recent progress on automatically searching augmentation policies has boosted the performance substantially for various tasks. A key component of automatic augmentation search is the evaluation process for a particular augmentation policy, which is utilized to return reward and usually runs thousands of times. A plain evaluation process, which includes full model training and validation, would be time-consuming. To achieve efficiency, many choose to sacrifice evaluation reliability for speed. In this paper, we dive into the dynamics of augmented training of the model. This inspires us to design a powerful and efficient proxy task based on the Augmentation-Wise Weight Sharing (AWS) to form a fast yet accurate evaluation process in an elegant way. Comprehensive analysis verifies the superiority of this approach in terms of effectiveness and efficiency. The augmentation policies found by our method achieve superior accuracies compared with existing auto-augmentation search methods. On CIFAR-10, we achieve a top-1 error rate of 1.24%, which is currently the best performing single model without extra training data. On ImageNet, we get a top-1 error rate of 20.36% for ResNet-50, which leads to 3.34% absolute error rate reduction over the baseline augmentation.

Estimating 3D hand pose from a single RGB image is a challenging task because of its ill-posed nature (i.e., depth ambiguity). Recently, various generative approaches have been proposed to predict the 3D joints of an RGB hand image by learning a unified latent space between two modalities (i.e., RGB image and 3D joints). However, projecting multi-modal data (i.e., RGB images and 3D joints) into a unified latent space is difficult as the modality-specific features usually interfere the learning of the optimal latent space. Hence in this paper, we propose to disentangle the latent space into two sub-latent spaces: modality- specific latent space and pose-specific latent space for 3D hand pose estimation. Our proposed method, namely Disentangled Cross-Modal Latent Space (DCMLS), consists of two variational autoencoder networks and auxiliary components which connect the two VAEs to align underlying hand poses and transfer modality-specific context from RGB to 3D. For the hand pose latent space, we align it with the two modalities by using a cross-modal discriminator with an adversarial learning strategy. For the context latent space, we learn a context translator to gain access to the cross-modal context. Experimental results on two widely used public benchmark datasets RHD and STB demonstrate that our proposed DCMLS method is able to clearly outperform the state-of-the-art ones on single image based 3D hand pose estimation.

Abstract Object detection, one of the most fundamental and challenging problems in computer vision, seeks to locate object instances from a large number of predefined categories in natural images. Deep learning techniques have emerged as a powerful strategy for learning feature representations directly from data and have led to remarkable breakthroughs in the field of generic object detection. Given this period of rapid evolution, the goal of this paper is to provide a comprehensive survey of the recent achievements in this field brought about by deep learning techniques. More than 300 research contributions are included in this survey, covering many aspects of generic object detection: detection frameworks, object feature representation, object proposal generation, context modeling, training strategies, and evaluation metrics. We finish the survey by identifying promising directions for future research.

In this work, we propose a division-and-summarization (DaS) framework for dense video captioning. After partitioning each untrimmed long video as multiple event proposals, where each event proposal consists of a set of short video segments, we extract visual feature (e.g., C3D feature) from each segment and use the existing image/video captioning approach to generate one sentence description for this segment. Considering that the generated sentences contain rich semantic descriptions about the whole event proposal, we formulate the dense video captioning task as a visual cue aided sentence summarization problem and propose a new two stage Long Short Term Memory (LSTM) approach equipped with a new hierarchical attention mechanism to summarize all generated sentences as one descriptive sentence with the aid of visual features. Specifically, the first-stage LSTM network takes all semantic words from the generated sentences and the visual features from all segments within one event proposal as the input, and acts as the encoder to effectively summarize both semantic and visual information related to this event proposal. The second-stage LSTM network takes the output from the first-stage LSTM network and the visual features from all video segments within one event proposal as the input, and acts as the decoder to generate one descriptive sentence for this event proposal. Our comprehensive experiments on the ActivityNet Captions dataset demonstrate the effectiveness of our newly proposed DaS framework for dense video captioning.

In this work, we tackle the problem of person search, which is a challenging task consisted of pedestrian detection and person re-identification (re-ID). Instead of sharing representations in a single joint model, we find that separating detector and re-ID feature extraction yields better performance. In order to extract more representative features for each identity, we segment out the foreground person from the original image patch. We propose a simple yet effective re-ID method, which models foreground person and original image patches individually, and obtains enriched representations from two separate CNN streams. We also propose a Confidence Weighted Stream Attention method which further re-adjusts the relative importance of the two streams by incorporating the detection confidence. Furthermore, we simplify the whole pipeline by incorporating semantic segmentation into the re-ID network, which is trained by bounding boxes as weakly-annotated masks and identification labels simultaneously. From the experiments on two standard person search benchmarks i.e. CUHK-SYSU and PRW, we achieve mAP of 83.3% and 32.8% respectively, surpassing the state of the art by a large margin. The extensive ablation study and model inspection further justifies our motivation.

Semantic attention has been shown to be effective in improving the performance of image captioning. The core of semantic attention based methods is to drive the model to attend to semantically important words, or attributes. In previous works, the attribute detector and the captioning network are usually independent, leading to the insufficient usage of the semantic information. Also, all the detected attributes, no matter whether they are appropriate for the linguistic context at the current step, are attended to through the whole caption generation process. This may sometimes disrupt the captioning model to attend to incorrect visual concepts. To solve these problems, we introduce two end-to-end trainable modules to closely couple attribute detection with image captioning as well as prompt the effective uses of attributes by predicting appropriate attributes at each time step. The multimodal attribute detector (MAD) module improves the attribute detection accuracy by using not only the image features but also the word embedding of attributes already existing in most captioning models. MAD models the similarity between the semantics of attributes and the image object features to facilitate accurate detection. The subsequent attribute predictor (SAP) module dynamically predicts a concise attribute subset at each time step to mitigate the diversity of image attributes. Compared to previous attribute based methods, our approach enhances the explainability in how the attributes affect the generated words and achieves a state-of-the-art single model performance of 128.8 CIDEr-D on the MSCOCO dataset. Extensive experiments on the MSCOCO dataset show that our proposal actually improves the performances in both image captioning and attribute detection simultaneously. The codes are available at: https://github.com/ RubickH/Image-Captioning-with-MAD-and-SAP.

Video compression algorithms are widely used to reduce the huge size of video data, but they also introduce unpleasant visual artifacts due to the lossy compression. In order to improve the quality of the compressed videos, we proposed a deep non-local Kalman network for compression artifact reduction. Specifically, the video restoration is modeled as a Kalman filtering procedure and the decoded frames can be restored from the proposed deep Kalman model. Instead of using the noisy previous decoded frames as temporal information, the less noisy previous restored frame is employed in a recursive way, which provides the potential to generate high quality restored frames. In the proposed framework, several deep neural networks are utilized to estimate the corresponding states in the Kalman filter and integrated together in the deep Kalman filtering network. More importantly, we also exploit the non-local prior information by incorporating the spatial and temporal non-local networks for better restoration. Our approach takes the advantages of both the model-based methods and learning-based methods, by combining the recursive nature of the Kalman model and powerful representation ability of neural networks. Extensive experimental results on the Vimeo-90k and HEVC benchmark datasets demonstrate the effectiveness of our proposed method.

Description-based person re-identification (Re-id) is an important task in video surveillance that requires discriminative cross-modal representations to distinguish different people. It is difficult to directly measure the similarity between images and descriptions due to the modality heterogeneity (the crossmodal problem). And all samples belonging to a single category (the fine-grained problem) makes this task even harder than the conventional image-description matching task. In this paper, we propose a Multi-granularity Image-text Alignments (MIA) model to alleviate the cross-modal fine-grained problem for better similarity evaluation in description-based person Re-id. Specifically, three different granularities, i.e., global-global, global-local and local-local alignments are carried out hierarchically. Firstly, the global-global alignment in the Global Contrast (GC) module is for matching the global contexts of images and descriptions. Secondly, the global-local alignment employs the potential relations between local components and global contexts to highlight the distinguishable components while eliminating the uninvolved ones adaptively in the Relation-guided Global-local Alignment (RGA) module. Thirdly, as for the local-local alignment, we match visual human parts with noun phrases in the Bi-directional Fine-grained Matching (BFM) module. The whole network combining multiple granularities can be end-to-end trained without complex preprocessing. To address the difficulties in training the combination of multiple granularities, an effective step training strategy is proposed to train these granularities step-by-step. Extensive experiments and analysis have shown that our method obtains the state-of-the-art performance on the CUHK-PEDES dataset and outperforms the previous methods by a significant margin.

Cascade is a classic yet powerful architecture that has boosted performance on various tasks. However, how to introduce cascade to instance segmentation remains an open question. A simple combination of Cascade R-CNN and Mask R-CNN only brings limited gain. In exploring a more effective approach, we find that the key to a successful instance segmentation cascade is to fully leverage the reciprocal relationship between detection and segmentation. In this work, we propose a new framework, Hybrid Task Cascade (HTC), which differs in two important aspects: (1) instead of performing cascaded refinement on these two tasks separately, it interweaves them for a joint multi-stage processing; (2) it adopts a fully convolutional branch to provide spatial context, which can help distinguishing hard foreground from cluttered background. Overall, this framework can learn more discriminative features progressively while integrating complementary features together in each stage. Without bells and whistles, a single HTC obtains 38.4% and 1.5% improvement over a strong Cascade Mask R-CNN baseline on MSCOCO dataset. Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge Object Detection Task. Code is available at https://github.com/open-mmlab/mmdetection.

We propose a unified framework for multi-person pose estimation and tracking. Our framework consists of two main components, i.e. SpatialNet and TemporalNet. The SpatialNet accomplishes body part detection and part-level data association in a single frame, while the TemporalNet groups human instances in consecutive frames into trajectories. Specifically, besides body part detection heatmaps, SpatialNet also predicts the Keypoint Embedding (KE) and Spatial Instance Embedding (SIE) for body part association. We model the grouping procedure into a differentiable Pose-Guided Grouping (PGG) module to make the whole part detection and grouping pipeline fully end-to-end trainable. TemporalNet extends the spatial grouping of keypoints to temporal grouping of human instances. Given human proposals from two consecutive frames, TemporalNet exploits both appearance features encoded in Human Embedding (HE) and temporally consistent geometric features embodied in Temporal Instance Embedding (TIE) for robust tracking. Extensive experiments demonstrate the effectiveness of our proposed model. Remarkably, we demonstrate substantial improvements over the state-of-the-art pose tracking method from 65.4% to 71.8% Multi-Object Tracking Accuracy (MOTA) on the ICCV'17 PoseTrack Dataset.

We present an efficient 3D object detection framework based on a single RGB image in the scenario of autonomous driving. Our efforts are put on extracting the underlying 3D information in a 2D image and determining the accurate 3D bounding box of object without point cloud or stereo data. Leveraging the off-the-shelf 2D object detector, we propose an artful approach to efficiently obtain a coarse cuboid for each predicted 2D box. The coarse cuboid has enough accuracy to guide us to determine the 3D box of the object by refinement. In contrast to previous state-of-the-art methods that only use the features extracted from the 2D bounding box for box refinement, we explore the 3D structure information of the object by employing the visual features of visible surfaces. The new features from surfaces are utilized to eliminate the problem of representation ambiguity brought by only using 2D bounding box. Moreover, we investigate different methods of 3D box refinement and discover that a classification formulation with quality aware loss have much better performance than regression. Evaluated on KITTI benchmark, our approach outperforms current state-of-the-art methods for single RGB image based 3D object detection.

Conventional video compression approaches use the predictive coding architecture and encode the corresponding motion information and residual information. In this paper, taking advantage of both classical architecture in the conventional video compression method and the powerful non-linear representation ability of neural networks, we propose the first end-to-end video compression deep model that jointly optimizes all the components for video compression. Specifically, learning based optical flow estimation is utilized to obtain the motion information and reconstruct the current frames. Then we employ two auto-encoder style neural networks to compress the corresponding motion and residual information. All the modules are jointly learned through a single loss function, in which they collaborate with each other by considering the trade-off between reducing the number of compression bits and improving quality of the decoded video. Experimental results show that the proposed approach can outperform the widely used video coding standard H.264 in terms of PSNR and be even on par with the latest standard H.265 in terms of MS-SSIM. Code is released at https://github.com/GuoLusjtu/DVC.

Compared with model architectures, the training process, which is also crucial to the success of detectors, has received relatively less attention in object detection. In this work, we carefully revisit the standard training practice of detectors, and find that the detection performance is often limited by the imbalance during the training process, which generally consists in three levels - sample level, feature level, and objective level. To mitigate the adverse effects caused thereby, we propose Libra R-CNN, a simple but effective framework towards balanced learning for object detection. It integrates three novel components: IoU-balanced sampling, balanced feature pyramid, and balanced L1 loss, respectively for reducing the imbalance at sample, feature, and objective level. Benefitted from the overall balanced design, Libra R-CNN significantly improves the detection performance. Without bells and whistles, it achieves 2.5 points and 2.0 points higher Average Precision (AP) than FPN Faster R-CNN and RetinaNet respectively on MSCOCO.

Semantic segmentation has achieved huge progress via adopting deep Fully Convolutional Networks (FCN). However, the performance of FCN based models severely rely on the amounts of pixel-level annotations which are expensive and time-consuming. To address this problem, it is a good choice to learn to segment with weak supervision from bounding boxes. How to make full use of the class-level and region-level supervisions from bounding boxes is the critical challenge for the weakly supervised learning task. In this paper, we first introduce a box-driven class-wise masking model (BCM) to remove irrelevant regions of each class. Moreover, based on the pixel-level segment proposal generated from the bounding box supervision, we could calculate the mean filling rates of each class to serve as an important prior cue, then we propose a filling rate guided adaptive loss (FR-Loss) to help the model ignore the wrongly labeled pixels in proposals. Unlike previous methods directly training models with the fixed individual segment proposals, our method can adjust the model learning with global statistical information. Thus it can help reduce the negative impacts from wrongly labeled proposals. We evaluate the proposed method on the challenging PASCAL VOC 2012 benchmark and compare with other methods. Extensive experimental results show that the proposed method is effective and achieves the state-of-the-art results.

Spatio-temporal action localization consists of three levels of tasks: spatial localization, action classification, and temporal segmentation. In this work, we propose a new Progressive Cross-stream Cooperation (PCSC) framework to iterative improve action localization results and generate better bounding boxes for one stream (i.e., Flow/RGB) by leveraging both region proposals and features from another stream (i.e., RGB/Flow) in an iterative fashion. Specifically, we first generate a larger set of region proposals by combining the latest region proposals from both streams, from which we can readily obtain a larger set of labelled training samples to help learn better action detection models. Second, we also propose a new message passing approach to pass information from one stream to another stream in order to learn better representations, which also leads to better action detection models. As a result, our iterative framework progressively improves action localization results at the frame level. To improve action localization results at the video level, we additionally propose a new strategy to train class-specific actionness detectors for better temporal segmentation, which can be readily learnt by using the training samples around temporal boundaries. Comprehensive experiments on two benchmark datasets UCF-101-24 and J-HMDB demonstrate the effectiveness of our newly proposed approaches for spatio-temporal action localization in realistic scenarios.

In crowd scenarios, reliable trajectory prediction of pedestrians requires insightful understanding of their social behaviors. These behaviors have been well investigated by plenty of studies, while it is hard to be fully expressed by hand-craft rules. Recent studies based on LSTM networks have shown great ability to learn social behaviors. However, many of these methods rely on previous neighboring hidden states but ignore the important current intention of the neighbors. In order to address this issue, we propose a data-driven state refinement module for LSTM network (SR-LSTM), which activates the utilization of the current intention of neighbors, and jointly and iteratively refines the current states of all participants in the crowd through a message passing mechanism. To effectively extract the social effect of neighbors, we further introduce a social-aware information selection mechanism consisting of an element-wise motion gate and a pedestrian-wise attention to select useful message from neighboring pedestrians. Experimental results on two public datasets, i.e. ETH and UCY, demonstrate the effectiveness of our proposed SR-LSTM and we achieve state-of-the-art results.

Human pose and shape are two important components of 2D human body. However, how to efficiently represent both of them in images is still an open question. In this paper, we propose the Triplet Representation for Body (TRB) --- a compact 2D human body representation, with skeleton keypoints capturing human pose information and contour keypoints containing human shape information. TRB not only preserves the flexibility of skeleton keypoint representation, but also contains rich pose and human shape information. Therefore, it promises broader application areas, such as human shape editing and conditional image generation. We further introduce the challenging problem of TRB estimation, where joint learning of human pose and shape is required. We construct several large-scale TRB estimation datasets, based on the popular 2D pose datasets LSP, MPII and COCO. To effectively solve TRB estimation, we propose a two-branch network (TRB-net) with three novel techniques, namely X-structure (Xs), Directional Convolution (DC) and Pairwise mapping (PM), to enforce multi-level message passing for joint feature learning. We evaluate our proposed TRB-net and several leading approaches on our proposed TRB datasets, and demonstrate the superiority of our method through extensive evaluations.

The fully-convolutional siamese network based on template matching has shown great potentials in visual tracking. During testing, the template is fixed with the initial target feature and the performance totally relies on the general matching ability of the siamese network. However, this manner cannot capture the temporal variations of targets or background clutter. In this work, we propose a novel gradient-guided network to exploit the discriminative information in gradients and update the template in the siamese network through feed-forward and backward operations. To be specific, the algorithm can utilize the information from the gradient to update the template in the current frame. In addition, a template generalization training method is proposed to better use gradient information and avoid overfitting. To our knowledge, this work is the first attempt to exploit the information in the gradient for template update in siamese-based trackers. Extensive experiments on recent benchmarks demonstrate that our method achieves better performance than other state-of-the-art trackers.

In this paper, we propose a level-aware progressive network (LAP-Net) for single image dehazing. Unlike previous multi-stage algorithms that generally learn in a coarse-to-fine fashion, each stage of LAP-Net learns different levels of haze with different supervision. Then the network can progressively learn the gradually aggravating haze. With this design, each stage can focus on a region with specific haze level and restore clear details. To effectively fuse the results of varying haze levels at different stages, we develop an adaptive integration strategy to yield the final dehazed image. This strategy is achieved by a hierarchical integration scheme, which is in cooperation with the memory network and the domain knowledge of dehazing to highlight the best-restored regions of each stage. Extensive experiments on both real-world images and two dehazing benchmarks validate the effectiveness of our proposed method.

Designing an effective loss function plays an important role in visual analysis. Most existing loss function designs rely on hand-crafted heuristics that require domain experts to explore the large design space, which is usually sub-optimal and time-consuming. In this paper, we propose AutoML for Loss Function Search (AM-LFS) which leverages REINFORCE to search loss functions during the training process. The key contribution of this work is the design of search space which can guarantee the generalization and transferability on different vision tasks by including a bunch of existing prevailing loss functions in a unified formulation. We also propose an efficient optimization framework which can dynamically optimize the parameters of loss function's distribution during training. Extensive experimental results on four benchmark datasets show that, without any tricks, our method outperforms existing hand-crafted loss functions in various computer vision tasks.

Data augmentation is critical to the success of modern deep learning techniques. In this paper, we propose Online Hyper-parameter Learning for Auto-Augmentation (OHL-Auto-Aug), an economical solution that learns the augmentation policy distribution along with network training. Unlike previous methods on auto-augmentation that search augmentation strategies in an offline manner, our method formulates the augmentation policy as a parameterized probability distribution, thus allowing its parameters to be optimized jointly with network parameters. Our proposed OHL-Auto-Aug eliminates the need of re-training and dramatically reduces the cost of the overall search process, while establishes significantly accuracy improvements over baseline models. On both CIFAR-10 and ImageNet, our method achieves remarkable on search accuracy, 60x faster on CIFAR-10 and 24x faster on ImageNet, while maintaining competitive accuracies.

Automatic estimation of the number of people in unconstrained crowded scenes is a challenging task and one major difficulty stems from the huge scale variation of people. In this paper, we propose a novel Deep Structured Scale Integration Network (DSSINet) for crowd counting, which addresses the scale variation of people by using structured feature representation learning and hierarchically structured loss function optimization. Unlike conventional methods which directly fuse multiple features with weighted average or concatenation, we first introduce a Structured Feature Enhancement Module based on conditional random fields (CRFs) to refine multiscale features mutually with a message passing mechanism. Specifically, each scale-specific feature is considered as a continuous random variable and passes complementary information to refine the features at other scales. Second, we utilize a Dilated Multiscale Structural Similarity loss to enforce our DSSINet to learn the local correlation of people's scales within regions of various size, thus yielding high-quality density maps. Extensive experiments on four challenging benchmarks well demonstrate the effectiveness of our method. In particular, our DSSINet achieves improvements of 9.5% error reduction on Shanghaitech dataset and 24.9% on UCF-QNRF dataset against the state-of-the-art methods.

In this paper, we propose a monocular 3D object detection framework in the domain of autonomous driving. Unlike previous image-based methods which focus on RGB feature extracted from 2D images, our method solves this problem in the reconstructed 3D space in order to exploit 3D contexts explicitly. To this end, we first leverage a stand-alone module to transform the input data from 2D image plane to 3D point clouds space for a better input representation, then we perform the 3D detection using PointNet backbone net to obtain objects' 3D locations, dimensions and orientations. To enhance the discriminative capability of point clouds, we propose a multi-modal feature fusion module to embed the complementary RGB cue into the generated point clouds representation. We argue that it is more effective to infer the 3D bounding boxes from the generated 3D scene space (i.e., X,Y, Z space) compared to the image plane (i.e., R,G,B image plane). Evaluation on the challenging KITTI dataset shows that our approach boosts the performance of state-of-the-art monocular approach by a large margin.

In this paper we tackle the joint learning problem of keyframe detection and visual odometry towards monocular visual SLAM systems. As an important task in visual SLAM, keyframe selection helps efficient camera relocalization and effective augmentation of visual odometry. To benefit from it, we first present a deep network design for the keyframe selection, which is able to reliably detect keyframes and localize new frames, then an end-to-end unsupervised deep framework further proposed for simultaneously learning the keyframe selection and the visual odometry tasks. As far as we know, it is the first work to jointly optimize these two complementary tasks in a single deep framework. To make the two tasks facilitate each other in the learning, a collaborative optimization loss based on both geometric and visual metrics is proposed. Extensive experiments on publicly available datasets (i.e. KITTI raw dataset and its odometry split [12]) clearly demonstrate the effectiveness of the proposed approach, and new state-ofthe-art results are established on the unsupervised depth and pose estimation from monocular video.

Recent saliency models extensively explore to incorporate multi-scale contextual information from Convolutional Neural Networks (CNNs). Besides direct fusion strategies, many approaches introduce message-passing to enhance CNN features or predictions. However, the messages are mainly transmitted in two ways, by feature-to-feature passing, and by prediction-to-prediction passing. In this paper, we add message-passing between features and predictions and propose a deep unified CRF saliency model . We design a novel cascade CRFs architecture with CNN to jointly refine deep features and predictions at each scale and progressively compute a final refined saliency map. We formulate the CRF graphical model that involves message-passing of feature-feature, feature-prediction, and prediction-prediction, from the coarse scale to the finer scale, to update the features and the corresponding predictions. Also, we formulate the mean-field updates for joint end-to-end model training with CNN through back propagation. The proposed deep unified CRF saliency model is evaluated over six datasets and shows highly competitive performance among the state of the arts.

Generative adversarial networks (GANs) have demonstrated great success in generating various visual content. However, images generated by existing GANs are often of attributes (e.g., smiling expression) learned from one image domain. As a result, generating images of multiple attributes requires many real samples possessing multiple attributes which are very resource expensive to be collected. In this paper, we propose a novel GAN, namely IntersectGAN, to learn multiple attributes from different image domains through an intersecting architecture. For example, given two image domains $X_1$ and $X_2$ with certain attributes, the intersection $X_1 \cap X_2$ denotes a new domain where images possess the attributes from both $X_1$ and $X_2$ domains. The proposed IntersectGAN consists of two discriminators $D_1$ and $D_2$ to distinguish between generated and real samples of different domains, and three generators where the intersection generator is trained against both discriminators. And an overall adversarial loss function is defined over three generators. As a result, our proposed IntersectGAN can be trained on multiple domains of which each presents one specific attribute, and eventually eliminates the need of real sample images simultaneously possessing multiple attributes. By using the CelebFaces Attributes dataset, our proposed IntersectGAN is able to produce high quality face images possessing multiple attributes (e.g., a face with black hair and a smiling expression). Both qualitative and quantitative evaluations are conducted to compare our proposed IntersectGAN with other baseline methods. Besides, several different applications of IntersectGAN have been explored with promising results.

Reconstructing 3D human shape and pose from a monocular image is challenging despite the promising results achieved by most recent learning based methods. The commonly occurred misalignment comes from the facts that the mapping from image to model space is highly non-linear and the rotation-based pose representation of the body model is prone to result in drift of joint positions. In this work, we present the Decompose-and-aggregate Network (DaNet) to address these issues. DaNet includes three new designs, namely UVI guided learning, decomposition for fine-grained perception, and aggregation for robust prediction. First, we adopt the UVI maps, which densely build a bridge between 2D pixels and 3D vertexes, as an intermediate representation to facilitate the learning of image-to-model mapping. Second, we decompose the prediction task into one global stream and multiple local streams so that the network not only provides global perception for the camera and shape prediction, but also has detailed perception for part pose prediction. Lastly, we aggregate the message from local streams to enhance the robustness of part pose prediction, where a position-aided rotation feature refinement strategy is proposed to exploit the spatial relationship between body parts. Such a refinement strategy is more efficient since the correlations between position features are stronger than that in the original rotation feature space. The effectiveness of our method is validated on the Human3.6M and UP-3D datasets. Experimental results show that the proposed method significantly improves the reconstruction performance in comparison with previous state-of-the-art methods. Our code is publicly available at https://github.com/HongwenZhang/DaNet-3DHumanReconstrution .

Referring expression comprehension, which locates the object instance described by a natural language expression, gains increasing interests in recent years. This paper aims at improving the task from two aspects: visual feature extraction and language features extraction. For visual feature extraction, we observe that most of the previous methods utilize only relative spatial information to model the visual relationship between object pairs while discarding rich semantic relationship between objects. This makes the visual-language matching difficult when the language expression contains semantic relationship to discriminate the referred object from other objects in the image. In this work, we propose a Semantic Visual Relationship Module (SVRM) to exploit this important information. For language feature extraction, a major problem comes from the long-tail distribution of words in the expressions. Since more than half of the words appear less than 20 times in the public datasets, deep models such as LSTM tend to fail to learn accurate representations for these words. To solve this problem, we propose a word2vec based word mapping method that maps these low frequency words to high frequency words with similar meaning. Experiments show that the proposed method outperforms existing state-of-the-art methods on three referring expression comprehension datasets.

Depth cues have been proved very useful in various computer vision and robotic tasks. This paper addresses the problem of monocular depth estimation from a single still image. Inspired by the effectiveness of recent works on multi-scale convolutional neural networks (CNN), we propose a deep model which fuses complementary information derived from multiple CNN side outputs. Different from previous methods using concatenation or weighted average schemes, the integration is obtained by means of continuous Conditional Random Fields (CRFs). In particular, we propose two different variations, one based on a cascade of multiple CRFs, the other on a unified graphical model. By designing a novel CNN implementation of mean-field updates for continuous CRFs, we show that both proposed models can be regarded as sequential deep networks and that training can be performed end-to-end. Through an extensive experimental evaluation, we demonstrate the effectiveness of the proposed approach and establish new state of the art results for the monocular depth estimation task on three publicly available datasets, i.e., NYUD-V2, Make3D and KITTI.

Online multi-object tracking (MOT) is a challenging problem and has many important applications including intelligence surveillance, robot navigation, and autonomous driving. In existing MOT methods, individual object's movements and inter-object relations are mostly modeled separately and relations between them are still manually tuned. In addition, inter-object relations are mostly modeled in a symmetric way, which we argue is not an optimal setting. To tackle those difficulties, in this paper, we propose a deep continuous conditional random field (DCCRF) for solving the online MOT problem in a track-by-detection framework. The DCCRF consists of unary and pairwise terms. The unary terms estimate tracked objects' displacements across time based on visual appearance information. They are modeled as deep convolution neural networks, which are able to learn discriminative visual features for tracklet association. The asymmetric pairwise terms model inter-object relations in an asymmetric way, which encourages high-confidence tracklets to help correct errors of low-confidence tracklets and not to be affected by low-confidence ones much. The DCCRF is trained in an end-to-end manner for better adapting the influences of visual information as well as inter-object relations. Extensive experimental comparisons with state-of-the-arts as well as detailed component analysis of our proposed DCCRF on two public benchmarks demonstrate the effectiveness of our proposed MOT framework.

Taxi demand prediction has recently attracted increasing research interest due to its huge potential application in large-scale intelligent transportation systems. However, most of the previous methods only considered the taxi demand prediction in origin regions, but neglected the modeling of the specific situation of the destination passengers. We believe it is suboptimal to preallocate the taxi into each region-based solely on the taxi origin demand. In this paper, we present a challenging and worth-exploring task, called taxi origin-destination demand prediction, which aims at predicting the taxi demand between all-region pairs in a future time interval. Its main challenges come from how to effectively capture the diverse contextual information to learn the demand patterns. We address this problem with a novel contextualized spatial-temporal network (CSTN), which consists of three components for the modeling of local spatial context (LSC), temporal evolution context (TEC), and global correlation context (GCC), respectively. First, an LSC module utilizes two convolution neural networks to learn the local spatial dependencies of taxi, demand respectively, from the origin view and the destination view. Second, a TEC module incorporates the local spatial features of taxi demand and the meteorological information to a Convolutional Long Short-term Memory Network (ConvLSTM) for the analysis of taxi demand evolution. Finally, a GCC module is applied to model the correlation between all regions by computing a global correlation feature as a weighted sum of all regional features, with the weights being calculated as the similarity between the corresponding region pairs. The extensive experiments and evaluations on a large-scale dataset well demonstrate the superiority of our CSTN over other compared methods for the taxi origin-destination demand prediction.

Depth estimation and scene parsing are two particularly important tasks in visual scene understanding. In this paper we tackle the problem of simultaneous depth estimation and scene parsing in a joint CNN. The task can be typically treated as a deep multi-task learning problem [42]. Different from previous methods directly optimizing multiple tasks given the input training data, this paper proposes a novel multi-task guided prediction-and-distillation network (PAD-Net), which first predicts a set of intermediate auxiliary tasks ranging from low level to high level, and then the predictions from these intermediate auxiliary tasks are utilized as multi-modal input via our proposed multi-modal distillation modules for the final tasks. During the joint learning, the intermediate tasks not only act as supervision for learning more robust deep representations but also provide rich multi-modal information for improving the final tasks. Extensive experiments are conducted on two challenging datasets (i.e. NYUD-v2 and Cityscapes) for both the depth estimation and scene parsing tasks, demonstrating the effectiveness of the proposed approach.

Recent advances in facial landmark detection achieve success by learning discriminative features from rich deformation of face shapes and poses. Besides the variance of faces themselves, the intrinsic variance of image styles, e.g., grayscale vs. color images, light vs. dark, intense vs. dull, and so on, has constantly been overlooked. This issue becomes inevitable as increasing web images are collected from various sources for training neural networks. In this work, we propose a style-aggregated approach to deal with the large intrinsic variance of image styles for facial landmark detection. Our method transforms original face images to style-aggregated images by a generative adversarial module. The proposed scheme uses the style-aggregated image to maintain face images that are more robust to environmental changes. Then the original face images accompanying with style-aggregated ones play a duet to train a landmark detector which is complementary to each other. In this way, for each face, our method takes two images as input, i.e., one in its original style and the other in the aggregated style. In experiments, we observe that the large variance of image styles would degenerate the performance of facial landmark detectors. Moreover, we show the robustness of our method to the large variance of image styles by comparing to a variant of our approach, in which the generative adversarial module is removed, and no style-aggregated images are used. Our approach is demonstrated to perform well when compared with state-of-the-art algorithms on benchmark datasets AFLW and 300-W. Code is publicly available on GitHub: https://github.com/D-X-Y/SAN.

Visual question answering (VQA) and visual question generation (VQG) are two trending topics in the computer vision, but they are usually explored separately despite their intrinsic complementary relationship. In this paper, we propose an end-to-end unified model, the Invertible Question Answering Network (iQAN), to introduce question generation as a dual task of question answering to improve the VQA performance. With our proposed invertible bilinear fusion module and parameter sharing scheme, our iQAN can accomplish VQA and its dual task VQG simultaneously. By jointly trained on two tasks with our proposed dual regularizes (termed as Dual Training), our model has a better understanding of the interactions among images, questions and answers. After training, iQAN can take either question or answer as input, and output the counterpart. Evaluated on the CLEVR and VQA2 datasets, our iQAN improves the top-1 accuracy of the prior art MUTAN VQA method by 1.33% and 0.88% (absolute increase) respectiely. We also show that our proposed dual training framework can consistently improve model performances of many popular VQA architectures1.

Person Re-identification (ReID) is an important yet challenging task in computer vision. Due to the diverse background clutters, variations on viewpoints and body poses, it is far from solved. How to extract discriminative and robust features invariant to background clutters is the core problem. In this paper, we first introduce the binary segmentation masks to construct synthetic RGB-Mask pairs as inputs, then we design a mask-guided contrastive attention model (MGCAM) to learn features separately from the body and background regions. Moreover, we propose a novel region-level triplet loss to restrain the features learnt from different regions, i.e., pulling the features from the full image and body region close, whereas pushing the features from backgrounds away. We may be the first one to successfully introduce the binary mask into person ReID task and the first one to propose region-level contrastive learning. We evaluate the proposed method on three public datasets, including MARS, Market-1501 and CUHK03. Extensive experimental results show that the proposed method is effective and achieves the state-of-the-art results. Mask and code will be released upon request.

Motion representation plays a vital role in human action recognition in videos. In this study, we introduce a novel compact motion representation for video action recognition, named Optical Flow guided Feature (OFF), which enables the network to distill temporal information through a fast and robust approach. The OFF is derived from the definition of optical flow and is orthogonal to the optical flow. The derivation also provides theoretical support for using the difference between two frames. By directly calculating pixel-wise spatio-temporal gradients of the deep feature maps, the OFF could be embedded in any existing CNN based video action recognition framework with only a slight additional cost. It enables the CNN to extract spatiotemporal information, especially the temporal information between frames simultaneously. This simple but powerful idea is validated by experimental results. The network with OFF fed only by RGB inputs achieves a competitive accuracy of 93.3% on UCF-101, which is comparable with the result obtained by two streams (RGB and optical flow), but is 15 times faster in speed. Experimental results also show that OFF is complementary to other motion modalities such as optical flow. When the proposed method is plugged into the state-of-the-art video action recognition framework, it has 96.0% and 74.2% accuracy on UCF-101 and HMDB-51 respectively. The code for this project is available at: https://github.com/kevin-ssy/Optical-Flow-Guided-Feature

We focus on the one-shot learning for video-based person re-Identification (re-ID). Unlabeled tracklets for the person re-ID tasks can be easily obtained by preprocessing, such as pedestrian detection and tracking. In this paper, we propose an approach to exploiting unlabeled tracklets by gradually but steadily improving the discriminative capability of the Convolutional Neural Network (CNN) feature representation via stepwise learning. We first initialize a CNN model using one labeled tracklet for each identity. Then we update the CNN model by the following two steps iteratively: 1. sample a few candidates with most reliable pseudo labels from unlabeled tracklets; 2. update the CNN model according to the selected data. Instead of the static sampling strategy applied in existing works, we propose a progressive sampling method to increase the number of the selected pseudo-labeled candidates step by step. We systematically investigate the way how we should select pseudo-labeled tracklets into the training set to make the best use of them. Notably, the rank-1 accuracy of our method outperforms the state-of-the-art method by 21.46 points (absolute, i.e., 62.67% vs. 41.21%) on the MARS dataset, and 16.53 points on the DukeMTMC-VideoReID dataset.

Person re-identification (ReID) is to identify pedestrians observed from different camera views based on visual appearance. It is a challenging task due to large pose variations, complex background clutters and severe occlusions. Recently, human pose estimation by predicting joint locations was largely improved in accuracy. It is reasonable to use pose estimation results for handling pose variations and background clutters, and such attempts have obtained great improvement in ReID performance. However, we argue that the pose information was not well utilized and hasn't yet been fully exploited for person ReID. In this work, we introduce a novel framework called Attention-Aware Compositional Network (AACN) for person ReID. AACN consists of two main components: Pose-guided Part Attention (PPA) and Attention-aware Feature Composition (AFC). PPA is learned and applied to mask out undesirable background features in pedestrian feature maps. Furthermore, pose-guided visibility scores are estimated for body parts to deal with part occlusion in the proposed AFC module. Extensive experiments with ablation analysis show the effectiveness of our method, and state-of-the-art results are achieved on several public datasets, including Market-1501, CUHK03, CUHK01, SenseReID, CUHK03-NP and DukeMTMC-reID.

Recently, remarkable advances have been achieved in 3D human pose estimation from monocular images because of the powerful Deep Convolutional Neural Networks (DCNNs). Despite their success on large-scale datasets collected in the constrained lab environment, it is difficult to obtain the 3D pose annotations for in-the-wild images. Therefore, 3D human pose estimation in the wild is still a challenge. In this paper, we propose an adversarial learning framework, which distills the 3D human pose structures learned from the fully annotated dataset to in-the-wild images with only 2D pose annotations. Instead of defining hard-coded rules to constrain the pose estimation results, we design a novel multi-source discriminator to distinguish the predicted 3D poses from the ground-truth, which helps to enforce the pose estimator to generate anthropometrically valid poses even with images in the wild. We also observe that a carefully designed information source for the discriminator is essential to boost the performance. Thus, we design a geometric descriptor, which computes the pairwise relative locations and distances between body joints, as a new information source for the discriminator. The efficacy of our adversarial learning framework with the new geometric descriptor has been demonstrated through extensive experiments on widely used public benchmarks. Our approach significantly improves the performance compared with previous state-of-the-art approaches.

In this paper, we propose a new unsupervised domain adaptation approach called Collaborative and Adversarial Network (CAN) through domain-collaborative and domain-adversarial training of neural networks. We add several domain classifiers on multiple CNN feature extraction blocks1, in which each domain classifier is connected to the hidden representations from one block and one loss function is defined based on the hidden presentation and the domain labels (e.g., source and target). We design a new loss function by integrating the losses from all blocks in order to learn domain informative representations from lower blocks through collaborative learning and learn domain uninformative representations from higher blocks through adversarial learning. We further extend our CAN method as Incremental CAN (iCAN), in which we iteratively select a set of pseudo-labelled target samples based on the image classifier and the last domain classifier from the previous training epoch and re-train our CAN model by using the enlarged training set. Comprehensive experiments on two benchmark datasets Office and ImageCLEF-DA clearly demonstrate the effectiveness of our newly proposed approaches CAN and iCAN for unsupervised domain adaptation.

Crowd counting from unconstrained scene images is a crucial task in many real-world applications like urban surveillance and management, but it is greatly challenged by the camera’s perspective that causes huge appearance variations in people’s scales and rotations. Conventional methods address such challenges by resorting to fixed multi-scale architectures that are often unable to cover the largely varied scales while ignoring the rotation variations. In this paper, we propose a unified neural network framework, named Deep Recurrent Spatial-Aware Network, which adaptively addresses the two issues in a learnable spatial transform module with a region-wise refinement process. Specifically, our framework incorporates a Recurrent Spatial-Aware Refinement (RSAR) module iteratively conducting two components: i) a Spatial Transformer Network that dynamically locates an attentional region from the crowd density map and transforms it to the suitable scale and rotation for optimal crowd estimation; ii) a Local Refinement Network that refines the density map of the attended region with residual learning. Extensive experiments on four challenging benchmarks show the effectiveness of our approach. Specifically, comparing with the existing best-performing methods, we achieve an improvement of 12\% on the largest dataset WorldExpo’10 and 22.8\% on the most challenging dataset UCF\_CC\_50

The basic principles in designing convolutional neural network (CNN) structures for predicting objects on different levels, e.g., image-level, region-level, and pixel-level, are diverging. Generally, network structures designed specifically for image classification are directly used as default backbone structure for other tasks including detection and segmentation, but there is seldom backbone structure designed under the consideration of unifying the advantages of networks designed for pixel-level or region-level predicting tasks, which may require very deep features with high resolution. Towards this goal, we design a fish-like network, called FishNet. In FishNet, the information of all resolutions is preserved and refined for the final task. Besides, we observe that existing works still cannot \emph{directly} propagate the gradient information from deep layers to shallow layers. Our design can better handle this problem. Extensive experiments have been conducted to demonstrate the remarkable performance of the FishNet. In particular, on ImageNet-1k, the accuracy of FishNet is able to surpass the performance of DenseNet and ResNet with fewer parameters. FishNet was applied as one of the modules in the winning entry of the COCO Detection 2018 challenge. The code is available at https://github.com/kevin-ssy/FishNet.

Feature extraction, deformation handling, occlusion handling, and classification are four important components in pedestrian detection. Existing methods learn or design these components either individually or sequentially. The interaction among these components is not yet well explored. This paper proposes that they should be jointly learned in order to maximize their strengths through cooperation. We formulate these four components into a joint deep learning framework and propose a new deep network architecture (Code available on www.ee.cuhk.edu.hk/wlouyang/projects/ouyangWiccv13Joint/index.html). By establishing automatic, mutual interaction among components, the deep model has average miss rate 8.57 percent/11.71 percent on the Caltech benchmark dataset with new/original annotations.

The visual cues from multiple support regions of different sizes and resolutions are complementary in classifying a candidate box in object detection. Effective integration of local and contextual visual cues from these regions has become a fundamental problem in object detection. In this paper, we propose a gated bi-directional CNN (GBD-Net) to pass messages among features from different support regions during both feature learning and feature extraction. Such message passing can be implemented through convolution between neighboring support regions in two directions and can be conducted in various layers. Therefore, local and contextual visual patterns can validate the existence of each other by learning their nonlinear relationships and their close interactions are modeled in a more complex way. It is also shown that message passing is not always helpful but dependent on individual samples. Gated functions are therefore needed to control message transmission, whose on-or-offs are controlled by extra visual evidence from the input sample. The effectiveness of GBD-Net is shown through experiments on three object detection datasets, ImageNet, Pascal VOC2007 and Microsoft COCO. Besides the GBD-Net, this paper also shows the details of our approach in winning the ImageNet object detection challenge of 2016, with source code provided on https://github.com/craftGBD/craftGBD. In this winning system, the modified GBD-Net, new pretraining scheme and better region proposal designs are provided. We also show the effectiveness of different network structures and existing techniques for object detection, such as multi-scale testing, left-right flip, bounding box voting, NMS, and context.

The state-of-the-art performance for object detection has been significantly improved over the past two years. Besides the introduction of powerful deep neural networks, such as GoogleNet and VGG, novel object detection frameworks, such as R-CNN and its successors, Fast R-CNN, and Faster R-CNN, play an essential role in improving the state of the art. Despite their effectiveness on still images, those frameworks are not specifically designed for object detection from videos. Temporal and contextual information of videos are not fully investigated and utilized. In this paper, we propose a deep learning framework that incorporates temporal and contextual information from tubelets obtained in videos, which dramatically improves the baseline performance of existing still-image detection frameworks when they are applied to videos. It is called T-CNN, i.e., tubelets with convolutional neueral networks. The proposed framework won newly introduced an object-detection-from-video task with provided data in the ImageNet Large-Scale Visual Recognition Challenge 2015. Code is publicly available athttps://github.com/myfavouritekk/T-CNN.

Pattern matching is widely used in signal processing, computer vision, and image and video processing. One efficient approach is to perform pattern matching in a transform domain that has good energy packing ability and so allows early rejection of most mismatched candidates. Calculating the transforms of pixels in sliding windows requires much computation, and so fast algorithms are employed. Existing methods require O(u) additions per pixel for projecting input pixels onto u 2D basis vectors. In this paper, we propose a new 2D transform, called asymmetric 2D Haar transform, and extend it to wavelet packets that contain exponentially large number of bases. A basis selection algorithm is then proposed to search for the optimal basis in the wavelet packets. A fast algorithm is also developed, which can compute u projection coefficients with only O(log u) additions per pixel. The results of experiments show that the proposed fast algorithm and the proposed transform can significantly accelerate the full-search-equivalent pattern matching process and outperform state-of-the-art methods.

In this paper, we propose to incorporate convolutional neural networks with a multi-context attention mechanism into an end-to-end framework for human pose estimation. We adopt stacked hourglass networks to generate attention maps from features at multiple resolutions with various semantics. The Conditional Random Field (CRF) is utilized to model the correlations among neighboring regions in the attention map. We further combine the holistic attention model, which focuses on the global consistency of the full human body, and the body part attention model, which focuses on detailed descriptions for different body parts. Hence our model has the ability to focus on different granularity from local salient regions to global semantic consistent spaces. Additionally, we design novel Hourglass Residual Units (HRUs) to increase the receptive field of the network. These units are extensions of residual units with a side branch incorporating filters with larger receptive field, hence features with various scales are learned and combined within the HRUs. The effectiveness of the proposed multi-context attention mechanism and the hourglass residual units is evaluated on two widely used human pose estimation benchmarks. Our approach outperforms all existing methods on both benchmarks over all the body parts. Code has been made publicly available.

Object detection in videos has drawn increasing attention recently with the introduction of the large-scale ImageNet VID dataset. Different from object detection in static images, temporal information in videos is vital for object detection. To fully utilize temporal information, state-of-the-art methods [15, 14] are based on spatiotemporal tubelets, which are essentially sequences of associated bounding boxes across time. However, the existing methods have major limitations in generating tubelets in terms of quality and efficiency. Motion-based [14] methods are able to obtain dense tubelets efficiently, but the lengths are generally only several frames, which is not optimal for incorporating long-term temporal information. Appearance-based [15] methods, usually involving generic object tracking, could generate long tubelets, but are usually computationally expensive. In this work, we propose a framework for object detection in videos, which consists of a novel tubelet proposal network to efficiently generate spatiotemporal proposals, and a Long Short-term Memory (LSTM) network that incorporates temporal information from tubelet proposals for achieving high object detection accuracy in videos. Experiments on the large-scale ImageNet VID dataset demonstrate the effectiveness of the proposed framework for object detection in videos.

As the intermediate level task connecting image captioning and object detection, visual relationship detection started to catch researchers attention because of its descriptive power and clear structure. It detects the objects and captures their pair-wise interactions with a subject-predicate-object triplet, e.g. person-ride-horse. In this paper, each visual relationship is considered as a phrase with three components. We formulate the visual relationship detection as three inter-connected recognition problems and propose a Visual Phrase guided Convolutional Neural Network (ViP-CNN) to address them simultaneously. In ViP-CNN, we present a Phrase-guided Message Passing Structure (PMPS) to establish the connection among relationship components and help the model consider the three problems jointly. Corresponding non-maximum suppression method and model training strategy are also proposed. Experimental results show that our ViP-CNN outperforms the state-of-art method both in speed and accuracy. We further pretrain ViP-CNN on our cleansed Visual Genome Relationship dataset, which is found to perform better than the pretraining on the ImageNet for this task.

This paper targets on the problem of set to set recognition, which learns the metric between two image sets. Images in each set belong to the same identity. Since images in a set can be complementary, they hopefully lead to higher accuracy in practical applications. However, the quality of each sample cannot be guaranteed, and samples with poor quality will hurt the metric. In this paper, the quality aware network (QAN) is proposed to confront this problem, where the quality of each sample can be automatically learned although such information is not explicitly provided in the training stage. The network has two branches, where the first branch extracts appearance feature embedding for each sample and the other branch predicts quality score for each sample. Features and quality scores of all samples in a set are then aggregated to generate the final feature embedding. We show that the two branches can be trained in an end-to-end manner given only the set-level identity annotation. Analysis on gradient spread of this mechanism indicates that the quality learned by the network is beneficial to set-to-set recognition and simplifies the distribution that the network needs to fit. Experiments on both face verification and person re-identification show advantages of the proposed QAN. The source code and network structure can be downloaded at GitHub.

This paper addresses the problem of depth estimation from a single still image. Inspired by recent works on multi-scale convolutional neural networks (CNN), we propose a deep model which fuses complementary information derived from multiple CNN side outputs. Different from previous methods, the integration is obtained by means of continuous Conditional Random Fields (CRFs). In particular, we propose two different variations, one based on a cascade of multiple CRFs, the other on a unified graphical model. By designing a novel CNN implementation of mean-field updates for continuous CRFs, we show that both proposed models can be regarded as sequential deep networks and that training can be performed end-to-end. Through extensive experimental evaluation we demonstrate the effectiveness of the proposed approach and establish new state of the art results on publicly available datasets.

Multi-label image classification is a fundamental but challenging task in computer vision. Great progress has been achieved by exploiting semantic relations between labels in recent years. However, conventional approaches are unable to model the underlying spatial relations between labels in multi-label images, because spatial annotations of the labels are generally not provided. In this paper, we propose a unified deep neural network that exploits both semantic and spatial relations between labels with only image-level supervisions. Given a multi-label image, our proposed Spatial Regularization Network (SRN) generates attention maps for all labels and captures the underlying relations between them via learnable convolutions. By aggregating the regularized classification results with original results by a ResNet-101 network, the classification performance can be consistently improved. The whole deep neural network is trained end-to-end with only image-level annotations, thus requires no additional efforts on image annotations. Extensive evaluations on 3 public datasets with different types of labels show that our approach significantly outperforms state-of-the-arts and has strong generalization capability. Analysis of the learned SRN model demonstrates that it can effectively capture both semantic and spatial relations of labels for improving classification performance.

In this paper, we propose a CNN-based framework for online MOT. This framework utilizes the merits of single object trackers in adapting appearance models and searching for target in the next frame. Simply applying single object tracker for MOT will encounter the problem in computational efficiency and drifted results caused by occlusion. Our framework achieves computational efficiency by sharing features and using ROI-Pooling to obtain individual features for each target. Some online learned target-specific CNN layers are used for adapting the appearance model for each target. In the framework, we introduce spatial-temporal attention mechanism (STAM) to handle the drift caused by occlusion and interaction among targets. The visibility map of the target is learned and used for inferring the spatial attention map. The spatial attention map is then applied to weight the features. Besides, the occlusion status can be estimated from the visibility map, which controls the online updating process via weighted loss on training samples with different occlusion statuses in different frames. It can be considered as temporal attention mechanism. The proposed algorithm achieves 34.3% and 46.0% in MOTA on challenging MOT15 and MOT16 benchmark dataset respectively.

Object detection, scene graph generation and region captioning, which are three scene understanding tasks at different semantic levels, are tied together: scene graphs are generated on top of objects detected in an image with their pairwise relationship predicted, while region captioning gives a language description of the objects, their attributes, relations and other context information. In this work, to leverage the mutual connections across semantic levels, we propose a novel neural network model, termed as Multi-level Scene Description Network (denoted as MSDN), to solve the three vision tasks jointly in an end-to-end manner. Object, phrase, and caption regions are first aligned with a dynamic graph based on their spatial and semantic connections. Then a feature refining structure is used to pass messages across the three levels of semantic tasks through the graph. We benchmark the learned model on three tasks, and show the joint learning across three tasks with our proposed method can bring mutual improvements over previous models. Particularly, on the scene graph generation task, our proposed method outperforms the stateof- art method with more than 3% margin. Code has been made publicly available.

Cascade is a widely used approach that rejects obvious negative samples at early stages for learning better classifier and faster inference. This paper presents chained cascade network (CC-Net). In this CC-Net, there are many cascade stages. Preceding cascade stages are placed at shallow layers. Easy hard examples are rejected at shallow layers so that the computation for deeper or wider layers is not required. In this way, features and classifiers at latter stages handle more difficult samples with the help of features and classifiers in previous stages. It yields consistent boost in detection performance on PASCAL VOC 2007 and ImageNet for both fast RCNN and Faster RCNN. CC-Net saves computation for both training and testing. Code is available on https://github.com/wk910930/ccnn.

Articulated human pose estimation is a fundamental yet challenging task in computer vision. The difficulty is particularly pronounced in scale variations of human body parts when camera view changes or severe foreshortening happens. Although pyramid methods are widely used to handle scale changes at inference time, learning feature pyramids in deep convolutional neural networks (DCNNs) is still not well explored. In this work, we design a Pyramid Residual Module (PRMs) to enhance the invariance in scales of DCNNs. Given input features, the PRMs learn convolutional filters on various scales of input features, which are obtained with different subsampling ratios in a multibranch network. Moreover, we observe that it is inappropriate to adopt existing methods to initialize the weights of multi-branch networks, which achieve superior performance than plain networks in many tasks recently. Therefore, we provide theoretic derivation to extend the current weight initialization scheme to multi-branch network structures. We investigate our method on two standard benchmarks for human pose estimation. Our approach obtains state-of-the-art results on both benchmarks. Code is available at https://github.com/bearpaw/PyraNet.

Recent works have shown that exploiting multi-scale representations deeply learned via convolutional neural networks (CNN) is of tremendous importance for accurate contour detection. This paper presents a novel approach for predicting contours which advances the state of the art in two fundamental aspects, i.e. multi-scale feature generation and fusion. Different from previous works directly considering multi-scale feature maps obtained from the inner layers of a primary CNN architecture, we introduce a hierarchical deep model which produces more rich and complementary representations. Furthermore, to refine and robustly fuse the representations learned at different scales, the novel Attention-Gated Conditional Random Fields (AG-CRFs) are proposed. The experiments ran on two publicly available datasets (BSDS500 and NYUDv2) demonstrate the effectiveness of the latent AG-CRF model and of the overall hierarchical framework.

In this paper, we propose deformable deep convolutional neural networks for generic object detection. This new deep learning object detection framework has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. A new pre-training strategy is proposed to learn feature representations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of model averaging. The proposed approach improves the mean averaged precision obtained by RCNN [16], which was the state-of-the-art, from 31% to 50.3% on the ILSVRC2014 detection test set. It also outperforms the winner of ILSVRC2014, GoogLeNet, by 6.1%. Detailed component-wise analysis is also provided through extensive experimental evaluation, which provides a global view for people to understand the deep learning object detection pipeline.

Human eyes can recognize person identities based on small salient regions, i.e., person saliency is distinctive and reliable in pedestrian matching across disjoint camera views. However, such valuable information is often hidden when computing similarities of pedestrian images with existing approaches. Inspired by our user study result of human perception on person saliency, we propose a novel perspective for person re-identification based on learning person saliency and matching saliency distribution. The proposed saliency learning and matching framework consists of four steps: (1) To handle misalignment caused by drastic viewpoint change and pose variations, we apply adjacency constrained patch matching to build dense correspondence between image pairs. (2) We propose two alternative methods, i.e., K-Nearest Neighbors and One-class SVM, to estimate a saliency score for each image patch, through which distinctive features stand out without using identity labels in the training procedure. (3) saliency matching is proposed based on patch matching. Matching patches with inconsistent saliency brings penalty, and images of the same identity are recognized by minimizing the saliency matching cost. (4) Furthermore, saliency matching is tightly integrated with patch matching in a unified structural RankSVM learning framework. The effectiveness of our approach is validated on the four public datasets. Our approach outperforms the state-of-the-art person re-identification methods on all these datasets.

In this paper, we propose a structured feature learning framework to reason the correlations among body joints at the feature level in human pose estimation. Different from existing approaches of modeling structures on score maps or predicted labels, feature maps preserve substantially richer descriptions of body joints. The relationships between feature maps of joints are captured with the introduced geometrical transform kernels, which can be easily implemented with a convolution layer. Features and their relationships are jointly learned in an end-to-end learning system. A bi-directional tree structured model is proposed, so that the feature channels at a body joint can well receive information from other joints. The proposed framework improves feature learning substantially. With very simple post processing, it reaches the best mean PCP on the LSP and FLIC datasets. Compared with the baseline of learning features at each joint separately with ConvNet, the mean PCP has been improved by 18% on FLIC. The code is released to the public.

Deep Convolution Neural Networks (CNNs) have shown impressive performance in various vision tasks such as image classification, object detection and semantic segmentation. For object detection, particularly in still images, the performance has been significantly increased last year thanks to powerful deep networks (e.g. GoogleNet) and detection frameworks (e.g. Regions with CNN features (RCNN)). The lately introduced ImageNet [6] task on object detection from video (VID) brings the object detection task into the video domain, in which objects' locations at each frame are required to be annotated with bounding boxes. In this work, we introduce a complete framework for the VID task based on still-image object detection and general object tracking. Their relations and contributions in the VID task are thoroughly studied and evaluated. In addition, a temporal convolution network is proposed to incorporate temporal information to regularize the detection results and shows its effectiveness for the task. Code is available at https://github.com/ myfavouritekk/vdetlib.

Finetuning from a pretrained deep model is found to yield state-of-the-art performance for many vision tasks. This paper investigates many factors that influence the performance in finetuning for object detection. There is a longtailed distribution of sample numbers for classes in object detection. Our analysis and empirical results show that classes with more samples have higher impact on the feature learning. And it is better to make the sample number more uniform across classes. Generic object detection can be considered as multiple equally important tasks. Detection of each class is a task. These classes/tasks have their individuality in discriminative visual appearance representation. Taking this individuality into account, we cluster objects into visually similar class groups and learn deep representations for these groups separately. A hierarchical feature learning scheme is proposed. In this scheme, the knowledge from the group with large number of classes is transferred for learning features in its subgroups. Finetuned on the GoogLeNet model, experimental results show 4.7% absolute mAP improvement of our approach on the ImageNet object detection dataset without increasing much computational cost at the testing stage. Code is available on www.ee.cuhk.edu.hk/ wlouyang/ projects/ImageNetFactors/CVPR16.html.

Due to the limited amount of training samples, finetuning pre-trained deep models online is prone to overfitting. In this paper, we propose a sequential training method for convolutional neural networks (CNNs) to effectively transfer pre-trained deep features for online applications. We regard a CNN as an ensemble with each channel of the output feature map as an individual base learner. Each base learner is trained using different loss criterions to reduce correlation and avoid over-training. To achieve the best ensemble online, all the base learners are sequentially sampled into the ensemble via important sampling. To further improve the robustness of each base learner, we propose to train the convolutional layers with random binary masks, which serves as a regularization to enforce each base learner to focus on different input features. The proposed online training method is applied to visual tracking problem by transferring deep features trained on massive annotated visual data and is shown to significantly improve tracking performance. Extensive experiments are conducted on two challenging benchmark data set and demonstrate that our tracking algorithm can outperform state-of-the-art methods with a considerable margin.

Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.

Recently, Deep Convolutional Neural Networks (DCNNs) have been applied to the task of human pose estimation, and have shown its potential of learning better feature representations and capturing contextual relationships. However, it is difficult to incorporate domain prior knowledge such as geometric relationships among body parts into DCNNs. In addition, training DCNN-based body part detectors without consideration of global body joint consistency introduces ambiguities, which increases the complexity of training. In this paper, we propose a novel end-to-end framework for human pose estimation that combines DCNNs with the expressive deformable mixture of parts. We explicitly incorporate domain prior knowledge into the framework, which greatly regularizes the learning process and enables the flexibility of our framework for loopy models or tree-structured models. The effectiveness of jointly learning a DCNN with a deformable mixture of parts model is evaluated through intensive experiments on several widely used benchmarks. The proposed approach significantly improves the performance compared with state-of-the-art approaches, especially on benchmarks with challenging articulations.

As a widely used non-linear activation, Rectified Linear Unit (ReLU) separates noise and signal in a feature map by learning a threshold or bias. However, we argue that the classification of noise and signal not only depends on the magnitude of responses, but also the context of how the feature responses would be used to detect more abstract patterns in higher layers. In order to output multiple response maps with magnitude in different ranges for a particular visual pattern, existing networks employing ReLU and its variants have to learn a large number of redundant filters. In this paper, we propose a multi-bias non-linear activation (MBA) layer to explore the information hidden in the magnitudes of responses. It is placed after the convolution layer to decouple the responses to a convolution kernel into multiple maps by multi-thresholding magnitudes, thus generating more patterns in the feature space at a low computational cost. It provides great flexibility of selecting responses to different visual patterns in different magnitude ranges to form rich representations in higher layers. Such a simple and yet effective scheme achieves the state-of-the-art performance on several benchmarks.

Deep convolutional neural networks (CNN) have achieved great success. On the other hand, modeling structural information has been proved critical in many vision problems. It is of great interest to integrate them effectively. In a classical neural network, there is no message passing between neurons in the same layer. In this paper, we propose a CRF-CNN framework which can simultaneously model structural information in both output and hidden feature layers in a probabilistic way, and it is applied to human pose estimation. A message passing scheme is proposed, so that in various layers each body joint receives messages from all the others in an efficient way. Such message passing can be implemented with convolution between features maps in the same layer, and it is also integrated with feedforward propagation in neural networks. Finally, a neural network implementation of end-to-end learning CRF-CNN is provided. Its effectiveness is demonstrated through experiments on two benchmark datasets.

Part-based models have demonstrated their merit in object detection. However, there is a key issue to be solved on how to integrate the inaccurate scores of part detectors when there are occlusions, abnormal deformations, appearances, or illuminations. To handle the imperfection of part detectors, this paper presents a probabilistic pedestrian detection framework. In this framework, a deformable part-based model is used to obtain the scores of part detectors and the visibilities of parts are modeled as hidden variables. Once the occluded parts are identified, their effects are properly removed from the final detection score. Unlike previous occlusion handling approaches that assumed independence among the visibility probabilities of parts or manually defined rules for the visibility relationship, a deep model is proposed in this paper for learning the visibility relationship among overlapping parts at multiple layers. The proposed approach can be viewed as a general postprocessing of part-detection results and can take detection scores of existing part-based models as input. The experimental results on three public datasets (Caltech, ETH, and Daimler) and a new CUHK occlusion dataset (http://www.ee.cuhk.edu.hk/~xgwang/CUHK_pedestrian.html), which is specially designed for the evaluation of occlusion handling approaches, show the effectiveness of the proposed approach.

In this paper, we propose deformable deep convolutional neural networks for generic object detection. This new deep learning object detection framework has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. A new pre-training strategy is proposed to learn feature representations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of model averaging. The proposed approach improves the mean averaged precision obtained by RCNN [14], which was the state-of-the-art, from 31% to 50.3% on the ILSVRC2014 detection test set. It also outperforms the winner of ILSVRC2014, GoogLeNet, by 6.1%. Detailed component-wise analysis is also provided through extensive experimental evaluation, which provide a global view for people to understand the deep learning object detection pipeline.

Low-level saliency cues or priors do not produce good enough saliency detection results especially when the salient object presents in a low-contrast background with confusing visual appearance. This issue raises a serious problem for conventional approaches. In this paper, we tackle this problem by proposing a multi-context deep learning framework for salient object detection. We employ deep Convolutional Neural Networks to model saliency of objects in images. Global context and local context are both taken into account, and are jointly modeled in a unified multi-context deep learning framework. To provide a better initialization for training the deep neural networks, we investigate different pre-training strategies, and a task-specific pre-training scheme is designed to make the multi-context modeling suited for saliency detection. Furthermore, recently proposed contemporary deep models in the ImageNet Image Classification Challenge are tested, and their effectiveness in saliency detection are investigated. Our approach is extensively evaluated on five public datasets, and experimental results show significant and consistent improvements over the state-of-the-art methods.

In this paper, we propose to predict immediacy for interacting persons from still images. A complete immediacy set includes interactions, relative distance, body leaning direction and standing orientation. These measures are found to be related to the attitude, social relationship, social interaction, action, nationality, and religion of the communicators. A large-scale dataset with 10,000 images is constructed, in which all the immediacy measures and the human poses are annotated. We propose a rich set of immediacy representations that help to predict immediacy from imperfect 1-person and 2-person pose estimation results. A multi-task deep recurrent neural network is constructed to take the proposed rich immediacy representation as input and learn the complex relationship among immediacy predictions multiple steps of refinement. The effectiveness of the proposed approach is proved through extensive experiments on the large scale dataset.

Learning strong feature representations from large scale supervision has achieved remarkable success in computer vision as the emergence of deep learning techniques. It is driven by big visual data with rich annotations. This paper contributes a large-scale object attribute database that contains rich attribute annotations (over 300 attributes) for ~180k samples and 494 object classes. Based on the ImageNet object detection dataset, it annotates the rotation, viewpoint, object part location, part occlusion, part existence, common attributes, and class-specific attributes. Then we use this dataset to train deep representations and extensively evaluate how these attributes are useful on the general object detection task. In order to make better use of the attribute annotations, a deep learning scheme is proposed by modeling the relationship of attributes and hierarchically clustering them into semantically meaningful mixture types. Experimental results show that the attributes are helpful in learning better features and improving the object detection accuracy by 2.6% in mAP on the ILSVRC 2014 object detection dataset and 2.4% in mAP on PASCAL VOC 2007 object detection dataset. Such improvement is well generalized across datasets.

We propose a new approach for general object tracking with fully convolutional neural network. Instead of treating convolutional neural network (CNN) as a black-box feature extractor, we conduct in-depth study on the properties of CNN features offline pre-trained on massive image data and classification task on ImageNet. The discoveries motivate the design of our tracking system. It is found that convolutional layers in different levels characterize the target from different perspectives. A top layer encodes more semantic features and serves as a category detector, while a lower layer carries more discriminative information and can better separate the target from distracters with similar appearance. Both layers are jointly used with a switch mechanism during tracking. It is also found that for a tracking target, only a subset of neurons are relevant. A feature map selection method is developed to remove noisy and irrelevant feature maps, which can reduce computation redundancy and improve tracking accuracy. Extensive evaluation on the widely used tracking benchmark [36] shows that the proposed tacker outperforms the state-of-the-art significantly.

In this paper, we address the challenging problem of detecting pedestrians who appear in groups. A new approach is proposed for single-pedestrian detection aided by two-pedestrian detection. A mixture model of two-pedestrian detectors is designed to capture the unique visual cues which are formed by nearby pedestrians but cannot be captured by single-pedestrian detectors. A probabilistic framework is proposed to model the relationship between the configurations estimated by single- and two-pedestrian detectors, and to refine the single-pedestrian detection result using two-pedestrian detection. The two-pedestrian detector can integrate with any single-pedestrian detector. Twenty-five state-of-the-art single-pedestrian detection approaches are combined with the two-pedestrian detector on three widely used public datasets: Caltech, TUD-Brussels, and ETH. Experimental results show that our framework improves all these approaches. The average improvement is 9 percent on the Caltech-Test dataset, 11 percent on the TUD-Brussels dataset and 17 percent on the ETH dataset in terms of average miss rate. The lowest average miss rate is reduced from 37 to percent on the Caltech-Test dataset, from 55 to 50 percent on the TUD-Brussels dataset and from 43 to 38 percent on the ETH dataset.

Visual appearance score, appearance mixture type and deformation are three important information sources for human pose estimation. This paper proposes to build a multi-source deep model in order to extract non-linear representation from these different aspects of information sources. With the deep model, the global, high-order human body articulation patterns in these information sources are extracted for pose estimation. The task for estimating body locations and the task for human detection are jointly learned using a unified deep model. The proposed approach can be viewed as a post-processing of pose estimation results and can flexibly integrate with existing methods by taking their information sources as input. By extracting the non-linear representation from multiple information sources, the deep model outperforms state-of-the-art by up to 8.6 percent on three public benchmark datasets.

In this paper, we propose a novel approach of learning mid-level filters from automatically discovered patch clusters for person re-identification. It is well motivated by our study on what are good filters for person re-identification. Our mid-level filters are discriminatively learned for identifying specific visual patterns and distinguishing persons, and have good cross-view invariance. First, local patches are qualitatively measured and classified with their discriminative power. Discriminative and representative patches are collected for filter learning. Second, patch clusters with coherent appearance are obtained by pruning hierarchical clustering trees, and a simple but effective cross-view training strategy is proposed to learn filters that are view-invariant and discriminative. Third, filter responses are integrated with patch matching scores in RankSVM training. The effectiveness of our approach is validated on the VIPeR dataset and the CUHK01 dataset. The learned mid-level features are complementary to existing handcrafted low-level features, and improve the best Rank-1 matching rate on the VIPeR dataset by 14%.

In this paper, we address the challenging problem of detecting pedestrians who appear in groups and have interaction. A new approach is proposed for single-pedestrian detection aided by multi-pedestrian detection. A mixture model of multi-pedestrian detectors is designed to capture the unique visual cues which are formed by nearby multiple pedestrians but cannot be captured by single-pedestrian detectors. A probabilistic framework is proposed to model the relationship between the configurations estimated by single-and multi-pedestrian detectors, and to refine the single-pedestrian detection result with multi-pedestrian detection. It can integrate with any single-pedestrian detector without significantly increasing the computation load. 15 state-of-the-art single-pedestrian detection approaches are investigated on three widely used public datasets: Caltech, TUD-Brussels and ETH. Experimental results show that our framework significantly improves all these approaches. The average improvement is 9% on the Caltech-Test dataset, 11% on the TUD-Brussels dataset and 17% on the ETH dataset in terms of average miss rate. The lowest average miss rate is reduced from 48% to 43% on the Caltech-Test dataset, from 55% to 50% on the TUD-Brussels dataset and from 51% to 41% on the ETH dataset.

Detecting pedestrians in cluttered scenes is a challenging problem in computer vision. The difficulty is added when several pedestrians overlap in images and occlude each other. We observe, however, that the occlusion/visibility statuses of overlapping pedestrians provide useful mutual relationship for visibility estimation - the visibility estimation of one pedestrian facilitates the visibility estimation of another. In this paper, we propose a mutual visibility deep model that jointly estimates the visibility statuses of overlapping pedestrians. The visibility relationship among pedestrians is learned from the deep model for recognizing co-existing pedestrians. Experimental results show that the mutual visibility deep model effectively improves the pedestrian detection results. Compared with existing image-based pedestrian detection approaches, our approach has the lowest average miss rate on the Caltech-Train dataset, the Caltech-Test dataset and the ETH dataset. Including mutual visibility leads to 4% - 8% improvements on multiple benchmark datasets.

Human eyes can recognize person identities based on some small salient regions. However, such valuable salient information is often hidden when computing similarities of images with existing approaches. Moreover, many existing approaches learn discriminative features and handle drastic viewpoint change in a supervised way and require labeling new training data for a different pair of camera views. In this paper, we propose a novel perspective for person re-identification based on unsupervised salience learning. Distinctive features are extracted without requiring identity labels in the training procedure. First, we apply adjacency constrained patch matching to build dense correspondence between image pairs, which shows effectiveness in handling misalignment caused by large viewpoint and pose variations. Second, we learn human salience in an unsupervised manner. To improve the performance of person re-identification, human salience is incorporated in patch matching to find reliable and discriminative matched patches. The effectiveness of our approach is validated on the widely used VIPeR dataset and ETHZ dataset.

Feature extraction, deformation handling, occlusion handling, and classification are four important components in pedestrian detection. Existing methods learn or design these components either individually or sequentially. The interaction among these components is not yet well explored. This paper proposes that they should be jointly learned in order to maximize their strengths through cooperation. We formulate these four components into a joint deep learning framework and propose a new deep network architecture. By establishing automatic, mutual interaction among components, the deep model achieves a 9% reduction in the average miss rate compared with the current best-performing pedestrian detection approaches on the largest Caltech benchmark dataset.

Cascaded classifiers have been widely used in pedestrian detection and achieved great success. These classifiers are trained sequentially without joint optimization. In this paper, we propose a new deep model that can jointly train multi-stage classifiers through several stages of back propagation. It keeps the score map output by a classifier within a local region and uses it as contextual information to support the decision at the next stage. Through a specific design of the training strategy, this deep architecture is able to simulate the cascaded classifiers by mining hard samples to train the network stage-by-stage. Each classifier handles samples at a different difficulty level. Unsupervised pre-training and specifically designed stage-wise supervised training are used to regularize the optimization problem. Both theoretical analysis and experimental results show that the training strategy helps to avoid over fitting. Experimental results on three datasets (Caltech, ETH and TUD-Brussels) show that our approach outperforms the state-of-the-art approaches.

Human salience is distinctive and reliable information in matching pedestrians across disjoint camera views. In this paper, we exploit the pair wise salience distribution relationship between pedestrian images, and solve the person re-identification problem by proposing a salience matching strategy. To handle the misalignment problem in pedestrian images, patch matching is adopted and patch salience is estimated. Matching patches with inconsistent salience brings penalty. Images of the same person are recognized by minimizing the salience matching cost. Furthermore, our salience matching is tightly integrated with patch matching in a unified structural Rank SVM learning framework. The effectiveness of our approach is validated on the VIPeR dataset and the CUHK Campus dataset. It outperforms the state-of-the-art methods on both datasets.

The gray-code kernels (GCK) family, which has Walsh Hadamard transform on sliding windows as a member, is a family of kernels that can perform image analysis efficiently using a fast algorithm, such as the GCK algorithm. The GCK has been successfully used for pattern matching. In this paper, we propose that the G4-GCK algorithm is more efficient than the previous algorithm in computing GCK. The G4-GCK algorithm requires four additions per pixel for three basis vectors independent of transform size and dimension. Based on the G4-GCK algorithm, we then propose the segmented GCK. By segmenting input data into L(s) parts, the SegGCK requires only four additions per pixel for 3L(s) basis vectors. Experimental results show that the proposed algorithm can significantly accelerate the full-search equivalent pattern matching process and outperforms state-of-the-art methods.

Part-based models have demonstrated their merit in object detection. However, there is a key issue to be solved on how to integrate the inaccurate scores of part detectors when there are occlusions or large deformations. To handle the imperfectness of part detectors, this paper presents a probabilistic pedestrian detection framework. In this framework, a deformable part-based model is used to obtain the scores of part detectors and the visibilities of parts are modeled as hidden variables. Unlike previous occlusion handling approaches that assume independence among visibility probabilities of parts or manually define rules for the visibility relationship, a discriminative deep model is used in this paper for learning the visibility relationship among overlapping parts at multiple layers. Experimental results on three public datasets (Caltech, ETH and Daimler) and a new CUHK occlusion dataset1specially designed for the evaluation of occlusion handling approaches show the effectiveness of the proposed approach.

Pattern matching is widely used in signal processing, computer vision, and image and video processing. Full search equivalent algorithms accelerate the pattern matching process and, in the meantime, yield exactly the same result as the full search. This paper proposes an analysis and comparison of state-of-the-art algorithms for full search equivalent pattern matching. Our intention is that the data sets and tests used in our evaluation will be a benchmark for testing future pattern matching algorithms, and that the analysis concerning state-of-the-art algorithms could inspire new fast algorithms. We also propose extensions of the evaluated algorithms and show that they outperform the original formulations.

Pattern matching is a widely used procedure in signal processing, computer vision, image and video processing. Recently, methods using Walsh Hadamard Transform (WHT) and Gray-Code kernels (GCK) are successfully applied for fast transform domain pattern matching. This paper introduces strip sum on the image. The sum of pixels in a rectangle can be computed by one addition using the strip sum. Then we propose to use the orthogonal Haar transform (OHT) for pattern matching. Applied for pattern matching, the algorithm using strip sum requires O(log u) additions per pixel to project input data of size N × N onto u 2-D OHT basis while existing fast algorithms require O(u) additions per pixel to project the same data onto u 2-D WHT or GCK basis. Experimental results show the efficiency of pattern matching using OHT.

This paper proposes a fast algorithm for Walsh Hadamard Transform on sliding windows which can be used to implement pattern matching most efficiently. The computational requirement of the proposed algorithm is about 1.5 additions per projection vector per sample, which is the lowest among existing fast algorithms for Walsh Hadamard Transform on sliding windows.

Blocking artifacts exist in images and video sequences compressed to low bit rates using block discrete cosine transform (DCT) compression standards. In order to reduce blocking artifacts, a novel DCT domain technique is presented in this paper. Firstly, a new FIR Wiener filter which exploits the dependence of neighboring DCT coefficients based on the linear minimum mean-square-error (LMMSE) criterion is proposed. Then we apply the new FIR Wiener filter twice in a dual adaptive filtering structure to restore each quantized DCT coefficient. In addition, an efficient parameter estimation method is proposed for the designed filter. Experimental results show that the performance of the proposed method is comparable to the state-of-the-art methods but has low computational complexity.

Edge detection plays an important role in digital image processing. Based on the non-decimated wavelet which is shift-invariant, in this paper, we develop a new edge detecting technique using 3-D Hidden Markov Model. Our proposed model can not only capture the relationship of the wavelet coefficients inter-scale, but also consider the intra-scale dependence. A computationally efficient maximum likelihood (ML) estimation algorithm is employed to compute parameters and the hidden state of each coefficient is revealed by maximum a posteriori (MAP) estimation. Experimental results of natural images are provided to evaluate the algorithm. In addition, the proposed model has the potential to be an efficient multi-scale statistical modeling tool for other image or video processing tasks.