Robotic manipulation requires precise spatial understanding to interact with objects in the real world. Point-based methods suffer from sparse sampling, leading to the loss of fine-grained semantics. Image-based methods typically feed RGB and depth into 2D backbones pre-trained on 3D auxiliary tasks, but their entangled semantics and geometry are sensitive to inherent depth noise in real-world that disrupts semantic understanding. Moreover, these methods focus on high-level geometry while overlooking low-level spatial cues essential for precise interaction. We propose SpatialActor, a disentangled framework for robust robotic manipulation that explicitly decouples semantics and geometry. The Semantic-guided Geometric Module adaptively fuses two complementary geometry from noisy depth and semantic-guided expert priors. Also, a Spatial Transformer leverages low-level spatial cues for accurate 2D-3D mapping and enables interaction among spatial features. We evaluate SpatialActor on multiple simulation and real-world scenarios across 50+ tasks. It achieves state-of-the-art performance with 87.4% on RLBench and improves by 13.9% to 19.4% under varying noisy conditions, showing strong robustness. Moreover, it significantly enhances few-shot generalization to new tasks and maintains robustness under various spatial perturbations.

Jingcheng Hu, Yinmin Zhang, Shijie Shang, Xiaobo Yang, Yue Peng, Zhewei Huang, Hebin Zhou, Xin Wu, Jie Cheng, Fanqi Wan, Xiangwen Kong, Chengyuan Yao, Kaiwen Yan, Ailin Huang, Hongyu Zhou, Qi Han, Zheng Ge, Xiangyu Zhang, Heung-Yeung Shum. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Xiangfeng Wang, Hangyu Guo, Yanlin Lai, Mitt Huang, Liang Zhao, Chengyuan Yao, Yinmin Zhang, Qi Han, Xiaoxiaoren, Chun Yuan, Tong Xu, Zheng Ge, Xiangyu Zhang, Daxin Jiang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Answering questions about charts presents a unique challenge for Vision-Language Models (VLMs). Unlike natural images, charts are structured artifacts governed by explicit visual grammar that demands pixel-level accuracy in visual perception. While recent VLMs demonstrate impressive reasoning abilities on chart tasks, a critical gap remains: their reasoning operates abstractly, disconnected from precise visual grounding. We introduce Active Perception, a framework that enables Gaze-Guided Thinking, a reasoning pattern that explicitly anchors abstract inference to concrete visual locations through coordinate-based operations (Locate, Trace, Extract, Compare). To instill this capability, we propose Skill Cultivation, a two-stage training strategy: Stage I injects coordinate-aware primitives via Supervised Fine-Tuning on ChartQAGaze-14K, our synthesized dataset of 14K coordinate-annotated reasoning chains; Stage II internalizes these skills into adaptive strategies via Reinforcement Learning with outcome-based rewards. Building upon Qwen2.5-VL-7B, Active Perception achieves state-of-the-art performance on ChartQA, improving overall accuracy from 78.96% to 82.44%, with particularly notable gains on the challenging Human split (75.76% to 81.28%). Qualitative analysis reveals emergent systematic chart-reading behaviors that mirror human visual strategies, demonstrating the effectiveness of spatially grounded reasoning for structured visual understanding. • We identify a critical gap in current VLMs for chart understanding: reasoning operates abstractly without precise visual grounding, limiting accurate data extraction from structured visualizations. • We propose Gaze-Guided Thinking , a reasoning pattern that anchors abstract inference to concrete visual locations through Coordinate Primitives ( Locate , Trace , Extract , Compare ), mimicking human chart scanning behavior. • We introduce Skill Cultivation , a two-stage training strategy combining SFT on ChartQAGaze-14K (14K coordinate-annotated reasoning chains) with outcome-based RL to inject and internalize spatially-grounded reasoning. • Active Perception achieves 82.44% overall accuracy on ChartQA (a 3.48% absolute improvement), with particularly strong performance on the Human split (81.28%, +5.52%), demonstrating that explicit visual grounding substantially enhances structured visual understanding. • Qualitative analysis reveals emergent human-like chart reading behaviors, where models systematically leverage coordinates for precise value extraction and spatial reasoning.

The field of autonomous driving increasingly demands high-quality annotated video training data. In this paper, we propose Panacea+, a powerful and universally applicable framework for generating video data in driving scenes. Built upon the foundation of our previous work, Panacea, Panacea+ adopts a multi-view appearance noise prior mechanism and a super-resolution module for enhanced consistency and increased resolution. Extensive experiments show that the generated video samples from Panacea+ greatly benefit a wide range of tasks on different datasets, including 3D object tracking, 3D object detection, and lane detection tasks on the nuScenes and Argoverse 2 dataset. These results strongly prove Panacea+ to be a valuable data generation framework for autonomous driving.

The widespread use of Large Language Models (LLMs) in software engineering has intensified the need for improved model and resource efficiency. In particular, for neural code generation, LLMs are used to translate function/method signature and DocString to executable code. DocStrings, which capture user requirements for the code and are typically used as the prompt for LLMs, often contain redundant information. Recent advancements in prompt compression have shown promising results in Natural Language Processing (NLP), but their applicability to code generation remains uncertain. Our empirical study shows that the state-of-the-art prompt compression methods achieve only about 10% reduction, as further reductions would cause significant performance degradation. In our study, we propose a novel compression method, ShortenDoc, dedicated to DocString compression for code generation. Our experiments on six code generation datasets, five open source LLMs (1B to 10B parameters), and one closed-source LLM GPT-4o confirm that ShortenDoc achieves 25–40% compression while preserving the quality of generated code, outperforming other baseline methods at similar compression levels. The benefit of this method is to improve efficiency and reduce the token processing cost while maintaining the quality of the generated code, especially when calling third-party APIs.

Code Language Models (CLMs), particularly those leveraging deep learning, have achieved significant success in code intelligence domain. However, the issue of security, particularly backdoor attacks, is often overlooked in this process. The previous research has focused on designing backdoor attacks for CLMs, but effective defenses have not been adequately addressed. In particular, existing defense methods from natural language processing, when directly applied to CLMs, are not effective enough and lack generality, working well in some models and scenarios but failing in others, thus fall short in consistently mitigating backdoor attacks. To bridge this gap, we first confirm the phenomenon of “early learning” as a general occurrence during the training of CLMs. This phenomenon refers to that a model initially focuses on the main features of training data but may become more sensitive to backdoor triggers over time, leading to overfitting and susceptibility to backdoor attacks. We then analyze that overfitting to backdoor triggers results from the use of the cross-entropy loss function, where the unboundedness of cross-entropy leads the model to increasingly concentrate on the features of the poisoned data. Based on this insight, we propose a general and effective loss function DeCE (Deceptive Cross-Entropy) by blending deceptive distributions and applying label smoothing to limit the gradient to bounded, which prevents the model from overfitting to backdoor triggers and then enhances the security of CLMs against backdoor attacks. To evaluate the effectiveness of our defense method, we select four code-related tasks as our experimental scenes and conduct experimental analyses on both natural language and two programming languages (Java and Python). Our experiments across multiple models with different sizes (from 125 millions to 7 billions) and poisoning ratios demonstrate the applicability and effectiveness of DeCE in enhancing the security of CLMs. The findings emphasize the potential of DeCE as a novel defense mechanism for CLMs, effectively tackling the challenge of securing models against backdoor threats.

Autonomous driving progress relies on large-scale annotated datasets. In this work, we explore the potential of generative models to produce vast quantities of freely-labeled data for autonomous driving applications and present SubjectDrive, the first model proven to scale generative data production in a way that could continuously improve autonomous driving applications. We investigate the impact of scaling up the quantity of generative data on the performance of downstream perception models and find that enhancing data diversity plays a crucial role in effectively scaling generative data production. Therefore, we have developed a novel model equipped with a subject control mechanism, which allows the generative model to leverage diverse external data sources for producing varied and useful data. Extensive evaluations confirm SubjectDrive's efficacy in generating scalable autonomous driving training data, marking a significant step toward revolutionizing data production methods in this field.

A new trend in the computer vision community is to capture objects of interest following flexible human command represented by a natural language prompt. However, the progress of using language prompts in driving scenarios is stuck in a bottleneck due to the scarcity of paired prompt-instance data. To address this challenge, we propose the first object-centric language prompt set for driving scenes within 3D, multi-view, and multi-frame space, named NuPrompt. It expands nuScenes dataset by constructing a total of 40,147 language descriptions, each referring to an average of 7.4 object tracklets. Based on the object-text pairs from the new benchmark, we formulate a novel prompt-based driving task, \ie, employing a language prompt to predict the described object trajectory across views and frames. Furthermore, we provide a simple end-to-end baseline model based on Transformer, named PromptTrack. Experiments show that our PromptTrack achieves impressive performance on NuPrompt. We hope this work can provide some new insights for the self-driving community.

The advent of large language models has significantly advanced automatic code generation, transforming the way programmers writing code. Inspired by natural language processing, mainstream code generation approaches represent code as a linear sequence of tokens. In this paper, we propose to represent code snippets as two-dimensional entities, where both code lines and tokens within lines are explicitly modeled. This representation allows us to capture the hierarchical and spatial structure of code, especially the dependencies between code lines. Our method CoDE introduces a dependency encoding approach that leverages dictionary learning to perform semantic matching between code lines. As such, it avoids the reliance on strict position indices, leading to better generalization to code with diverse context and lengths. We thoroughly evaluate CoDE based on four categories of tasks. The experimental results showcase its generalizability, context understanding and retrieval, as well as interpretability in code generation.

The success of large language models has driven interest in developing similar speech processing capabilities. However, a key challenge is the scarcity of high-quality spontaneous speech data, as most existing datasets contain scripted dialogues. To address this, we present a novel pipeline for eliciting and recording natural dialogues and release our dataset with 100+ hours of spontaneous speech. Our approach fosters fluid, natural conversations while encouraging a diverse range of topics and interactive exchanges. Unlike traditional methods, it facilitates genuine interactions, providing a reproducible framework for future data collection. This paper introduces our dataset and methodology, laying the groundwork for addressing the shortage of spontaneous speech data. We plan to expand this dataset in future stages, offering a growing resource for the research community.

We introduce MAGI, a hybrid video generation framework that combines masked modeling for intra-frame generation with causal modeling for next-frame generation. Our key innovation, Complete Teacher Forcing (CTF), conditions masked frames on complete observation frames rather than masked ones (namely Masked Teacher Forcing, MTF), enabling a smooth transition from token-level (patch-level) to frame-level autoregressive generation. CTF significantly outperforms MTF, achieving a +23% improvement in FVD scores on first-frame conditioned video prediction. To address issues like exposure bias, we employ targeted training strategies, setting a new benchmark in autoregressive video generation. Experiments show that MAGI can generate long, coherent video sequences exceeding 100 frames, even when trained on as few as 16 frames, highlighting its potential for scalable, high-quality video generation.

This paper describes our best-performing submission for the ICASSP-2025 Signal Processing Grand Challenge PROCESS, focused on the classification of speech into 3 groups - Healthy, Mild Cognitive Impairment (MCI), and Dementia - using three speech tasks in English. Our approach was aligned with the aim of simple, preclinical detection of dementia, employing a minimal set of acoustic features, and no linguistic analysis. We built an ensemble classifier based on 1) Selected features from the ComParE acoustic feature set and 2) knowledge-based rules for combining predictions across the 3 tasks, using a two-tier majority vote system. Our technique outperformed the baseline results by a large margin, achieving a macro-F1 of 0.96 on the development set, 0.99 on 5-fold cross-validation and 0.64 on the test set.

The Mamba-based model has demonstrated outstanding performance across tasks in computer vision, natural language processing, and speech processing. However, in the realm of speech processing, the Mamba-based model’s performance varies across different tasks. For instance, in tasks such as speech enhancement and spectrum reconstruction, the Mamba model performs well when used independently. However, for tasks like speech recognition, additional modules are required to surpass the performance of attention-based models. We propose the hypothesis that the Mamba-based model excels in "reconstruction" tasks within speech processing. However, for "classification tasks" such as Speech Recognition, additional modules are necessary to accomplish the "reconstruction" step. To validate our hypothesis, we analyze the previous Mamba-based Speech Models from an information theory perspective. Furthermore, we leveraged the properties of HuBERT in our study. We trained a Mamba-based HuBERT model, and the mutual information patterns, along with the model’s performance metrics, confirmed our assumptions.

Personalized image generation holds great promise in assisting humans in everyday work and life due to its impressive function in creatively generating personalized content. However, current evaluations either are automated but misalign with humans or require human evaluations that are time-consuming and expensive. In this work, we present DreamBench++, a human-aligned benchmark that advanced multimodal GPT models automate. Specifically, we systematically design the prompts to let GPT be both human-aligned and self-aligned, empowered with task reinforcement. Further, we construct a comprehensive dataset comprising diverse images and prompts. By benchmarking 7 modern generative models, we demonstrate that \dreambench results in significantly more human-aligned evaluation, helping boost the community with innovative findings.

This paper introduces reconstructive visual instruction tuning (ROSS), a family of Large Multimodal Models (LMMs) that exploit vision-centric supervision signals. In contrast to conventional visual instruction tuning approaches that exclusively supervise text outputs, ROSS prompts LMMs to supervise visual outputs via reconstructing input images. By doing so, it capitalizes on the inherent richness and detail present within input images themselves, which are often lost in pure text supervision. However, producing meaningful feedback from natural images is challenging due to the heavy spatial redundancy of visual signals. To address this issue, ROSS employs a denoising objective to reconstruct latent representations of input images, avoiding directly regressing exact raw RGB values. This intrinsic activation design inherently encourages LMMs to maintain image detail, thereby enhancing their fine-grained comprehension capabilities and reducing hallucinations. Empirically, ROSS consistently brings significant improvements across different visual encoders and language models. In comparison with extrinsic assistance state-of-the-art alternatives that aggregate multiple visual experts, ROSS delivers competitive performance with a single SigLIP visual encoder, demonstrating the efficacy of our vision-centric supervision tailored for visual outputs. The code will be made publicly available upon acceptance.

The generation and simulation of diverse real-world scenes have significant application value in the field of autonomous driving, especially for the corner cases. Recently, researchers have explored employing neural radiance fields or diffusion models to generate novel views or synthetic data under driving scenes. However, these approaches suffer from unseen scenes or restricted video length, thus lacking sufficient adaptability for data generation and simulation. To address these issues, we propose a simple yet effective framework, named Glad, to generate video data in a frame-by-frame style. To ensure the temporal consistency of synthetic video, we introduce a latent variable propagation module, which views the latent features of previous frame as noise prior and injects it into the latent features of current frame. In addition, we design a streaming data sampler to orderly sample the original image in a video clip at continuous iterations. Given the reference frame, our Glad can be viewed as a streaming simulator by generating the videos for specific scenes. Extensive experiments are performed on the widely-used nuScenes dataset. Experimental results demonstrate that our proposed Glad achieves promising performance, serving as a strong baseline for online video generation. We will release the source code and models publicly.

In the pursuit of superior video-processing MLLMs, we have encountered a perplexing paradox: the “anti-scaling law”, where more data and larger models lead to worse performance. This study unmasks the culprit: “temporal hacking”, a phenomenon where models shortcut by fixating on select frames, missing the full video narrative. In this work, we systematically establish a comprehensive theory of temporal hacking, defining it from a reinforcement learning perspective, introducing the Temporal Perplexity (TPL) score to assess this misalignment, and proposing the Unhackable Temporal Rewarding (UTR) framework to mitigate the temporal hacking. Both theoretically and empirically, TPL proves to be a reliable indicator of temporal modeling quality, correlating strongly with frame activation patterns. Extensive experiments reveal that UTR not only counters temporal hacking but significantly elevates video comprehension capabilities. This work not only advances video-AI systems but also illuminates the critical importance of aligning proxy rewards with true objectives in MLLM development.

We present a perception in reflection paradigm designed to transcend the limitations of current large vision-language models (LVLMs), which are expected yet often fail to achieve perfect perception initially. Specifically, we propose Reflective Perception (RePer), a dual-model reflection mechanism that systematically alternates between policy and critic models, enables iterative refinement of visual perception. This framework is powered by Reflective Perceptual Learning (RPL), which reinforces intrinsic reflective capabilities through a methodically constructed visual reflection dataset and reflective unlikelihood training Comprehensive experimental evaluation demonstrates RePer's quantifiable improvements in image understanding, captioning precision, and hallucination reduction. Notably, RePer achieves strong alignment between model attention patterns and human visual focus, while RPL optimizes fine-grained and free-form preference alignment. These advancements establish perception in reflection as a robust paradigm for future multimodal agents, particularly in tasks requiring complex reasoning and multi-step manipulation. Project Page: [https://weiyana.github.io/Perception-in-Reflection](https://weiyana.github.io/Perception-in-Reflection)

We introduce Open-Reasoner-Zero, the first open source implementation of large-scale reasoning-oriented RL training on the base model focusing on scalability, simplicity and accessibility. Through extensive experiments, we demonstrate that a minimalist approach, vanilla PPO with GAE ($\lambda=1$, $\gamma=1$) and straightforward rule-based rewards, without any KL regularization, is sufficient to scale up both benchmark performance and response length, replicating the scaling phenomenon observed in DeepSeek-R1-Zero. Using the same base model as DeepSeek-R1-Zero-Qwen-32B, our implementation achieves superior performance across AIME2024, MATH500, and GPQA Diamond, while demonstrating remarkable efficiency—requiring only 1/10 of the training steps compared to the DeepSeek-R1-Zero pipeline. We validate that this recipe generalizes well across diverse training domains and different model families without algorithmic modifications. Moreover, our analysis not only covers training dynamics and ablation for critical design choices, but also quantitatively show how the learned critic in Reasoner-Zero training effectively identifies and devalues repetitive response patterns, yielding more robust advantage estimations and enhancing training stability. Embracing the principles of open-source, we release our source code, parameter settings, training data, and model weights across various sizes, fostering reproducibility and encouraging further exploration of the properties of related models.

Training Large Language Models (LLMs) is prohibitively expensive, creating a critical scaling gap where insights from small-scale experiments often fail to transfer to resource-intensive production systems, thereby hindering efficient innovation. To bridge this, we introduce Farseer, a novel and refined scaling law offering enhanced predictive accuracy across scales. By systematically constructing a model loss surface $L(N,D)$, Farseer achieves a significantly better fit to empirical data than prior laws (e.g., \Chinchilla's law). Our methodology yields accurate, robust, and highly generalizable predictions, demonstrating excellent extrapolation capabilities, outperforming Chinchilla's law, whose extrapolation error is 433\% higher. This allows for the reliable evaluation of competing training strategies across all $(N,D)$ settings, enabling conclusions from small-scale ablation studies to be confidently extrapolated to predict large-scale performance. Furthermore, Farseer provides new insights into optimal compute allocation, better reflecting the nuanced demands of modern LLM training. To validate our approach, we trained an extensive suite of approximately 1,000 LLMs across diverse scales and configurations, consuming roughly 3 million NVIDIA H100 GPU hours. To foster further research, we are comprehensively open-sourcing all code, data, results (https://github.com/Farseer-Scaling-Law/Farseer), all training logs (https://wandb.ai/billzid/Farseer?nw=nwuserbillzid), all models used in scaling law fitting (https://huggingface.co/Farseer-Scaling-Law).

The remarkable reasoning capability of large language models (LLMs) stems from cognitive behaviors that emerge through reinforcement with verifiable rewards. This work investigates how to transfer this principle to Multimodal LLMs (MLLMs) to unlock advanced visual reasoning. We introduce a two-stage paradigm built on Qwen2.5-VL-7B: a massive linguistic cold-start fine-tuning, followed by multimodal reinforcement learning (RL) spanning nearly 1,000 steps—surpassing all previous open-source efforts in scale. This pioneering work reveals three fundamental insights: 1) Behavior transfer emerges surprisingly early in cold start due to linguistic mental imagery. 2) Cold start broadly memorizes visual behaviors, while RL critically discerns and scales up effective patterns. 3) Transfer strategically favors high-utility behaviors such as visual reflection. Our resulting model, Open-Vision-Reasoner (OVR), achieves state-of-the-art performance on a suite of reasoning benchmarks, including 95.3% on MATH500, 51.8% on MathVision and 54.6% on MathVerse. We release our model, data, and training dynamics to catalyze the development of more capable, behavior-aligned multimodal reasoners.

With the rapid development of Large Vision Language Models, the focus of Graphical User Interface (GUI) agent tasks shifts from single-screen tasks to complex screen navigation challenges. However, real-world GUI environments, such as PC software and mobile Apps, are often complex and proprietary, making it difficult to obtain the comprehensive environment information needed for agent training and evaluation. This limitation hinders systematic investigation and benchmarking of agent navigation capabilities. To address this limitation, we introduce GUI Exploration Lab, a simulation environment engine for GUI agent navigation research that enables flexible definition and composition of screens, icons, and navigation graphs, while providing full access to environment information for comprehensive agent training and evaluation. Through extensive experiments, we find that supervised fine-tuning enables effective memorization of fundamental knowledge, serving as a crucial foundation for subsequent training. Building on this, single-turn reinforcement learning further enhances generalization to unseen scenarios. Finally, multi-turn reinforcement learning encourages the development of exploration strategies through interactive trial and error, leading to further improvements in screen navigation performance. We validate our methods on both static and interactive benchmarks, demonstrating that our findings generalize effectively to real-world scenarios. These findings demonstrate the advantages of reinforcement learning approaches in GUI navigation and offer practical guidance for building more capable and generalizable GUI agents.

Inspired by the success of DeepSeek-R1, we explore the potential of rule-based reinforcement learning (RL) in MLLM post-training for perception policy learning. While promising, our initial experiments reveal that incorporating a thinking process through RL does not consistently lead to performance gains across all visual perception tasks. This leads us to delve into the essential role of RL in the context of visual perception. In this work, we return to the fundamentals and explore the effects of RL on different perception tasks. We observe that the perceptual perplexity is a major factor in determining the effectiveness of RL. We also observe that reward design plays a crucial role in further approaching the upper limit of model perception. To leverage these findings, we propose Perception-R1, a scalable RL framework using GRPO during MLLM post-training. With a standard Qwen2-VL-2B-Instruct, Perception-R1 achieves +4.2% on RefCOCO+, +17.9% on PixMo-Count, +4.2% on PageOCR, and notably, 31.9% AP on COCO2017 val for the first time, establishing a strong baseline for perception policy learning.

In this work, we present a novel direction to build an image tokenizer directly on top of a frozen vision foundation model, which is a largely underexplored area. Specifically, we employ a frozen vision foundation model as the encoder of our tokenizer. To enhance its effectiveness, we introduce two key components: (1) a region-adaptive quantization framework that reduces redundancy in the pre-trained features on regular 2D grids, and (2) a semantic reconstruction objective that aligns the tokenizer’s outputs with the foundation model’s representations to preserve semantic fidelity. Based on these designs, our proposed image tokenizer, \textbf{\ours}, achieves substantial improvements in image reconstruction and generation quality, while also enhancing token efficiency. It further boosts autoregressive (AR) generation---achieving a gFID of \textbf{1.36} on ImageNet benchmarks, while accelerating model convergence by \textbf{three times}, and enabling high-fidelity class-conditional synthesis without the need for classifier-free guidance (CFG). The code is available at \href{https://github.com/CVMI-Lab/VFMTok}{https://github.com/CVMI-Lab/VFMTok}.

With the widespread application of Large Language Models (LLMs) in the field of natural language processing and software engineering, security vulnerabilities have emerged as a critical concern. Among these, jailbreak attacks represent a prevalent security threat, as they bypass the internal security checks of the model through carefully designed input perturbations, generating malicious outputs which severely may compromise the reliability and security of the LLM-based software tools. Existing defense methods based on reinforcement learning and fine-tuning often suffer from limited generalization, low interpretability, and high computational overhead. To address these limitations, we propose MaskedDefender, a novel defense approach that detects potential attack features by analyzing model's response differences to various inputs. Guided by the principle of key information protection and selective compression, MaskedDefender identifies critical tokens associated with jailbreak attacks by optimizing the gradient of a multi-objective loss function. It then applies soft guidance to steer the model's attention toward these critical tokens. Our approach highlights jailbreak intentions and reduces the model's confusion in identifying such attacks without modifying model parameters. Experimental results show that MaskedDefender outperforms existing defense methods in enabling the model to detect and resist jailbreak attacks, while maintaining both efficiency and effectiveness.

Masked visual modeling has attracted much attention due to its promising potential in learning generalizable representations. Typical approaches urge models to predict specific contents of masked tokens, which can be intuitively considered as teaching a student (the model) to solve given problems (predicting masked contents). Under such settings, the performance is highly correlated with mask strategies (the difficulty of provided problems). We argue that it is equally important for the model to stand in the shoes of a teacher to produce challenging problems by itself. Intuitively, patches with high values of reconstruction loss can be regarded as hard samples, and masking those hard patches naturally becomes a demanding reconstruction task. To empower the model as a teacher, we propose Hard Patch Mining (HPM), predicting patch-wise losses and subsequently determining where to mask. Technically, we introduce an auxiliary loss predictor, which is trained with a relative objective to prevent overfitting to exact loss values. To gradually guide the training procedure, we propose an easy-to-hard mask strategy. Empirically, HPM brings significant improvements under both image and video benchmarks. Interestingly, solely incorporating the extra loss prediction objective leads to better representations, verifying the efficacy of determining where is hard to reconstruct.

The development of large language models (LLMs) has revolutionized automated code generation. However, their high demand of computation resources has hindered a broader deployment and raised environmental concerns. A common strategy for diminishing computational demands is to cache Key-Value (KV) states from the attention mechanism which is adopted predominately by mainstream LLMs. It can mitigate the need of repeated attention computations, but brings significant memory overhead. Current practices in NLP often use sparse attention which may, unfortunately, lead to substantial inaccuracies, or hallucinations, in code generation tasks. In this paper, we analyze the attention weights distribution within code generation models via an empirical study, uncovering a sparsity pattern, i.e., the aggregation of information at specific anchor points. Based on this observation, we propose a novel approach, AnchorCoder, which features token-wise anchor attention designed to extract and compress the contextual information, and layer-wise anchor attention enabling cross-layer communication to mitigate the issue of excessive superposition caused by the compression. The extensive experiments across multiple benchmark datasets confirm the effectiveness of AnchorCoder, which can consistently achieve a significant (at least 70%) reduction in KV cache requirements, while preserving the majority of model’s performance.

Recently, masked image modeling (MIM) has demonstrated promising prospects in self-supervised representation learning. However, existing MIM frameworks recover all masked patches equivalently, ignoring that the reconstruction difficulty of different patches can vary sharply due to their diverse distance from visible patches. In this paper, we propose a novel deep dynamic supervision to enable MIM methods to dynamically reconstruct patches with different degrees of difficulty at different pretraining phases and depths of the model. Our deep dynamic supervision helps to provide more locality inductive bias for ViTs especially in deep layers, which inherently makes up for the absence of local prior for self-attention mechanism. Built upon the deep dynamic supervision, we propose Deep Dynamic AutoEncoder (DDAE), a simple yet effective MIM framework that utilizes dynamic mechanisms for pixel regression and feature self-distillation simultaneously. Extensive experiments across a variety of vision tasks including ImageNet classification, semantic segmentation on ADE20K and object detection on COCO demonstrate the effectiveness of our approach.

Recently 3D object detection from surround-view images has made notable advancements with its low deployment cost. However, most works have primarily focused on close perception range while leaving long-range detection less explored. Expanding existing methods directly to cover long distances poses challenges such as heavy computation costs and unstable convergence. To address these limitations, this paper proposes a novel sparse query-based framework, dubbed Far3D. By utilizing high-quality 2D object priors, we generate 3D adaptive queries that complement the 3D global queries. To efficiently capture discriminative features across different views and scales for long-range objects, we introduce a perspective-aware aggregation module. Additionally, we propose a range-modulated 3D denoising approach to address query error propagation and mitigate convergence issues in long-range tasks. Significantly, Far3D demonstrates SoTA performance on the challenging Argoverse 2 dataset, covering a wide range of 150 meters, surpassing several LiDAR-based approaches. The code is available at https://github.com/megvii-research/Far3D.

Vision-Language Models (VLMs) such as CLIP have demonstrated remarkable generalization capabilities to downstream tasks. However, existing prompt tuning based frameworks need to parallelize learnable textual inputs for all categories, suffering from massive GPU memory consumption when there is a large number of categories in the target dataset. Moreover, previous works require to include category names within prompts, exhibiting subpar performance when dealing with ambiguous category names. To address these shortcomings, we propose Compound Text-Guided Prompt Tuning (TGP-T) that significantly reduces resource demand while achieving superior performance. We introduce text supervision to the optimization of prompts, which enables two benefits: 1) releasing the model reliance on the pre-defined category names during inference, thereby enabling more flexible prompt generation; 2) reducing the number of inputs to the text encoder, which decreases GPU memory consumption significantly. Specifically, we found that compound text supervisions, i.e., category-wise and content-wise, is highly effective, since they provide inter-class separability and capture intra-class variations, respectively. Moreover, we condition the prompt generation on visual features through a module called Bonder, which facilitates the alignment between prompts and visual features. Extensive experiments on few-shot recognition and domain generalization demonstrate that TGP-T achieves superior performance with consistently lower training costs. It reduces GPU memory usage by 93% and attains a 2.5% performance gain on 16-shot ImageNet. The code is available at https://github.com/EricTan7/TGP-T.

The field of autonomous driving increasingly demands high-quality annotated training data. In this paper, we propose Panacea, an innovative approach to generate panoramic and controllable videos in driving scenarios, capable of yielding an unlimited numbers of diverse, annotated samples pivotal for autonomous driving advancements. Panacea addresses two critical challenges: ‘Consistency’ and ‘Controllability.’ Consistency ensures temporal and cross-view coherence, while Controllability ensures the alignment of generated content with corresponding annotations. Our approach integrates a novel 4D attention and a two-stage generation pipeline to maintain coherence, supplemented by the ControlNet framework for meticulous control by the Bird'View (BEV) layouts. Extensive qualitative and quantitative evaluations of Panacea on the nuScenes dataset prove its effectiveness in generating high-quality multi-view driving-scene videos. This work notably propels the field of autonomous driving by effectively augmenting the training dataset used for advanced BEV perception techniques.

Xiangyu Zhang, Daijiao Liu, Hexin Liu, Qiquan Zhang, Hanyu Meng, Leibny Paola Garcia Perera, EngSiong Chng, Lina Yao. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Depression is a critical concern in global mental health, prompting extensive research into AIbased detection methods.Among various AI technologies, Large Language Models (LLMs) stand out for their versatility in mental healthcare applications.However, their primary limitation arises from their exclusive dependence on textual input, which constrains their overall capabilities.Furthermore, the utilization of LLMs in identifying and analyzing depressive states is still relatively untapped.In this paper, we present an innovative approach to integrating acoustic speech information into the LLMs framework for multimodal depression detection.We investigate an efficient method for depression detection by integrating speech signals into LLMs utilizing Acoustic Landmarks.By incorporating acoustic landmarks, which are specific to the pronunciation of spoken words, our method adds critical dimensions to text transcripts.This integration also provides insights into the unique speech patterns of individuals, revealing the potential mental states of individuals.Evaluations of the proposed approach on the DAIC-WOZ dataset reveal state-of-the-art results when compared with existing Audio-Text baselines.In addition, this approach is not only valuable for the detection of depression but also represents a new perspective in enhancing the ability of LLMs to comprehend and process speech signals.

While significant advancements in artificial intelligence (AI) have catalyzed progress across various domains, its full potential in understanding visual perception remains underexplored. We propose an artificial neural network dubbed VISION, an acronym for "Visual Interface System for Imaging Output of Neural activity," to mimic the human brain and show how it can foster neuroscientific inquiries. Using visual and contextual inputs, this multimodal model predicts the brain's functional magnetic resonance imaging (fMRI) scan response to natural images. VISION successfully predicts human hemodynamic responses as fMRI voxel values to visual inputs with an accuracy exceeding state-of-the-art performance by 45%. We further probe the trained networks to reveal representational biases in different visual areas, generate experimentally testable hypotheses, and formulate an interpretable metric to associate these hypotheses with cortical functions. With both a model and evaluation metric, the cost and time burdens associated with designing and implementing functional analysis on the visual cortex could be reduced. Our work suggests that the evolution of computational models may shed light on our fundamental understanding of the visual cortex and provide a viable approach toward reliable brain-machine interfaces.

Languages usually switch within a multilingual speech signal, especially in a bilingual society. This phenomenon is referred to as code-switching (CS), making automatic speech recognition (ASR) challenging under a multilingual scenario. We propose to improve CS-ASR by biasing the hybrid CTC/attention ASR model with multi-level language information comprising frame-and token-level language posteriors. The interaction between various resolutions of language biases is subsequently explored in this work. We conducted experiments on datasets from the ASRU 2019 code-switching challenge. Compared to the baseline, the proposed interactive language biases (ILB) method achieves higher performance and ablation studies highlight the effects of different language biases and their interactions. In addition, the results presented indicate that language bias implicitly enhances internal language modeling, leading to performance degradation after employing an external language model.

This paper presents DreamLLM, a learning framework that first achieves versatile Multimodal Large Language Models (MLLMs) empowered with frequently overlooked synergy between multimodal comprehension and creation. DreamLLM operates on two fundamental principles. The first focuses on the generative modeling of both language and image posteriors by direct sampling in the raw multimodal space. This approach circumvents the limitations and information loss inherent to external feature extractors like CLIP, and a more thorough multimodal understanding is obtained. Second, DreamLLM fosters the generation of raw, interleaved documents, modeling both text and image contents, along with unstructured layouts. This allows DreamLLM to learn all conditional, marginal, and joint multimodal distributions effectively. As a result, DreamLLM is the first MLLM capable of generating free-form interleaved content. Comprehensive experiments highlight DreamLLM's superior performance as a zero-shot multimodal generalist, reaping from the enhanced learning synergy. Project page: https://dreamllm.github.io.

This paper makes the first attempt towards unsupervised preference alignment in Vision-Language Models (VLMs). We generate chosen and rejected responses with regard to the original and augmented image pairs, and conduct preference alignment with direct preference optimization. It is based on a core idea: properly designed augmentation to the image input will induce VLM to generate false but hard negative responses, which helps the model to learn from and produce more robust and powerful answers. The whole pipeline no longer hinges on supervision from GPT-4 or human involvement during alignment, and is highly efficient with few lines of code. With only 8k randomly sampled unsupervised data, it achieves 90% relative score to GPT-4 on complex reasoning in LLaVA-Bench, and improves LLaVA-7B/13B by 6.7%/5.6% score on complex multi-modal benchmark MM-Vet. Visualizations shows its improved ability to align with user-intentions. A series of ablations are firmly conducted to reveal the latent mechanism of the approach, which also indicates its potential towards further scaling.

Large Language Models (LLMs) have demonstrated remarkable potential in code generation. The integration of Chain of Thought (CoT) reasoning can further boost their performance. However, current CoT methods often require manual writing or LLMs with over 100 billion parameters to generate, impeding their applicability in resource-constrained scenarios. In this study, we investigate lightweight Language Models ($\ell$LMs), which are defined to have fewer than 10 billion parameters. Empirically, we find that most$\ell$LMs cannot generate high-quality CoTs when prompted by the few-shot method, but can take advantage of high-quality CoTs generated elsewhere to improve their performance in code generation. Based on these findings, we design a novel approachCOTTONwhich can leverage$\ell$LMs to automatically generate CoTs for code generation. We synthesize new datasets and conduct extensive experiments on various benchmarks. The results show that the CoTs generated byCOTTONoutperform the baselines in terms of automated and human evaluation metrics. In particular, the CoTs generated byCOTTONboost various$\ell$LMs to achieve higher performance gains than those generated by LLMs such as ChatGLM (130B), and are competitive with those generated by Gemini and gpt-3.5-turbo. The results also reveal thatCOTTONnot only improves the performance of$\ell$LMs, but also enhances the performance of LLMs. Our study showcases the potential of$\ell$LMs in software engineering applications.

Recent advance in 2D CNNs has revealed that large kernels are important. However, when directly applying large convolutional kernels in 3D CNNs, severe difficulties are met, where those successful module designs in 2D become surprisingly ineffective on 3D networks, including the popular depth-wise convolution. To address this vital challenge, we instead propose the spatial-wise partition convolution and its large-kernel module. As a result, it avoids the optimization and efficiency issues of naive 3D large kernels. Our large-kernel 3D CNN network, LargeKernel3D, yields notable improvement in 3D tasks of semantic segmentation and object detection. It achieves 73.9% mIoU on the ScanNetv2 semantic segmentation and 72.8% NDS nuScenes object detection benchmarks, ranking 1st on the nuScenes LIDAR leaderboard. The performance further boosts to 74.2% NDS with a simple multi-modal fusion. In addition, LargeKernel3D can be scaled to 17×17×17 kernel size on Waymo 3D object detection. For the first time, we show that large kernels are feasible and essential for 3D visual tasks. Our code and models is available at github.com/dvlab-research/LargeKernel3D.

3D object detectors usually rely on hand-crafted proxies, e.g., anchors or centers, and translate well-studied 2D frameworks to 3D. Thus, sparse voxel features need to be densified and processed by dense prediction heads, which inevitably costs extra computation. In this paper, we instead propose VoxelNext for fully sparse 3D object detection. Our core insight is to predict objects directly based on sparse voxel features, without relying on hand-crafted proxies. Our strong sparse convolutional network VoxelNeXt detects and tracks 3D objects through voxel features entirely. It is an elegant and efficient framework, with no need for sparse-to-dense conversion or NMS post-processing. Our method achieves a better speed-accuracy trade-off than other mainframe detectors on the nuScenes dataset. For the first time, we show that a fully sparse voxel-based representation works decently for LIDAR 3D object detection and tracking. Extensive experiments on nuScenes, Waymo, and Argoverse2 benchmarks validate the effectiveness of our approach. Without bells and whistles, our model outperforms all existing LIDAR methods on the nuScenes tracking test benchmark. Code and models are available at github.com/dvlab-research/VoxelNeXt.

Recently, Masked Image Modeling (MIM) achieves great success in self-supervised visual recognition. However, as a reconstruction-based framework, it is still an open question to understand how MIM works, since MIM appears very different from previous well-studied siamese approaches such as contrastive learning. In this paper, we propose a new viewpoint: MIM implicitly learns occlusion-invariant features, which is analogous to other siamese methods while the latter learns other invariance. By relaxing MIM formulation into an equivalent siamese form, MIM methods can be interpreted in a unified framework with conventional methods, among which only a) data transformations, i.e. what invariance to learn, and b) similarity measurements are different. Furthermore, taking MAE [29] as a representative example of MIM, we empirically find the success of MIM models relates a little to the choice of similarity functions, but the learned occlusion invariant feature introduced by masked image - it turns out to be a favored initialization for vision transformers, even though the learned feature could be less semantic. We hope our findings could inspire researchers to develop more powerful self-supervised methods in computer vision community.

Existing referring understanding tasks tend to involve the detection of a single text-referred object. In this paper, we propose a new and general referring understanding task, termed referring multi-object tracking (RMOT). Its core idea is to employ a language expression as a semantic cue to guide the prediction of multi-object tracking. To the best of our knowledge, it is the first work to achieve an arbitrary number of referent object predictions in videos. To push forward RMOT, we construct one benchmark with scalable expressions based on KITTI, named Refer-KITTI. Specifically, it provides 18 videos with 818 expressions, and each expression in a video is annotated with an average of 10.7 objects. Further, we develop a transformer-based architecture TransRMOT to tackle the new task in an online manner, which achieves impressive detection performance and out-performs other counterparts. The Refer-KITTI dataset and the code are released at https://referringmot.github.io.

Differentiable architecture search (DARTS) has signif-icantly promoted the development of NAS techniques because of its high search efficiency and effectiveness but suf-fers from performance collapse. In this paper, we make efforts to alleviate the performance collapse problem for DARTS from two aspects. First, we investigate the expres-sive power of the supernet in DARTS and then derive a new setup of DARTS paradigm with only training Batch-Norm. Second, we theoretically find that random features dilute the auxiliary connection role of skip-connection in supernet optimization and enable search algorithm focus on fairer operation selection, thereby solving the performance collapse problem. We instantiate DARTS and PC-DARTS with random features to build an improved version for each named RF-DARTS and RF-PCDARTS respectively. Experimental results show that RF-DARTS obtains 94.36% test accuracy on CIFAR-10 (which is the nearest optimal result in NAS-Bench-201), and achieves the newest state-of-the-art top-1 test error of 24.0% on ImageNet when transferring from CIFAR-10. Moreover, RF-DARTS performs robustly across three datasets (CIFAR-10, CIFAR-100, and SVHN) and four search spaces (S1-S4). Besides, RF-PCDARTS achieves even better results on ImageNet, that is, 23.9% top-1 and 7.1% top-5 test error, surpassing representative methods like single-path, training-free, and partial-channel paradigms directly searched on ImageNet.

In this paper, we propose MOTRv2, a simple yet effective pipeline to bootstrap end-to-end multi-object tracking with a pretrained object detector. Existing end-to-end methods, e.g. MOTR [43] and TrackFormer [20] are inferior to their tracking-by-detection counterparts mainly due to their poor detection performance. We aim to improve MOTR by elegantly incorporating an extra object detector. We first adopt the anchor formulation of queries and then use an extra object detector to generate proposals as anchors, providing detection prior to MOTR. The simple modification greatly eases the conflict between joint learning detection and association tasks in MOTR. MOTRv2 keeps the query propogation feature and scales well on large-scale benchmarks. MOTRv2 ranks the 1st place (73.4% HOTA on DanceTrack) in the 1st Multiple People Tracking in Group Dance Challenge. Moreover, MOTRv2 reaches state-of-the-art performance on the BDD100K dataset. We hope this simple and effective pipeline can provide some new insights to the end-to-end MOT community. Code is available at https://github.com/megvii-research/MOTRv2.

A recent study has shown a phenomenon called neural collapse in that the within-class means of features and the classifier weight vectors converge to the vertices of a simplex equiangular tight frame at the terminal phase of training for classification. In this paper, we explore the cor-responding structures of the last-layer feature centers and classifiers in semantic segmentation. Based on our empirical and theoretical analysis, we point out that semantic segmentation naturally brings contextual correlation and imbalanced distribution among classes, which breaks the equiangular and maximally separated structure of neural collapse for both feature centers and classifiers. However, such a symmetric structure is beneficial to discrimination for the minor classes. To preserve these advantages, we in-troduce a regularizer on feature centers to encourage the network to learn features closer to the appealing structure in imbalanced semantic segmentation. Experimental results show that our method can bring significant improvements on both 2D and 3D semantic segmentation bench-marks. Moreover, our method ranks 1stand sets a new record (+6.8% mIoU) on the ScanNet200 test leaderboard.

With careful manipulation, malicious agents can reverse engineer private information encoded in pre-trained language models. Security concerns motivate the development of quantum pre-training. In this work, we propose a highly portable quantum language model (PQLM) that can easily transmit information to downstream tasks on classical machines. The framework consists of a cloud PQLM built with random Variational Quantum Classifiers (VQC) and local models for downstream applications. We demonstrate the ad hoc portability of the quantum model by extracting only the word embeddings and effectively applying them to downstream tasks on classical machines. Our PQLM exhibits comparable performance to its classical counterpart on both intrinsic evaluation (loss, perplexity) and extrinsic evaluation (multilingual sentiment analysis accuracy) metrics. We also perform ablation studies on the factors affecting PQLM performance to analyze model stability. Our work establishes a theoretical foundation for a portable quantum pre-trained language model that could be trained on private data and made available for public use with privacy protection guarantees.

The detection of abnormal fetal heartbeats during pregnancy is important for monitoring the health conditions of the fetus. While adult ECG has made several advances in modern medicine, noninvasive fetal electrocardiography (FECG) remains a great challenge. In this paper, we introduce a new method based on affine combinations of adaptive filters to extract FECG signals. The affine combination of multiple filters is able to precisely fit the reference signal, and thus obtain more accurate FECGs. We proposed a method to combine the Least Mean Square (LMS) and Recursive Least Squares (RLS) filters. Our approach found that the Combined Recursive Least Squares (CRLS) filter achieves the best performance among all proposed combinations. In addition, we found that CRLS is more advantageous in extracting FECG from abdominal electrocardiograms (AECG) with a small signal-to-noise ratio (SNR). Compared with the state-of-the-art Multiple Sub-Filter Adaptive Noise Canceller (MSF-ANC) method, CRLS shows improved performance. The sensitivity, accuracy and F1 score are improved by 3.58%, 2.39% and 1.36%, respectively.

In this paper, we propose PETRv2, a unified framework for 3D perception from multi-view images. Based on PETR [25], PETRv2 explores the effectiveness of temporal modeling, which utilizes the temporal information of previous frames to boost 3D object detection. More specifically, we extend the 3D position embedding (3D PE) in PETR for temporal modeling. The 3D PE achieves the temporal alignment on object position of different frames. To support for multi-task learning (e.g., BEV segmentation and 3D lane detection), PETRv2 provides a simple yet effective solution by introducing task-specific queries, which are initialized under different spaces. PETRv2 achieves state-of-the-art performance on 3D object detection, BEV segmentation and 3D lane detection. Detailed robustness analysis is also conducted on PETR framework. Code is available at https://github.com/megvii-research/PETR.

In this paper, we propose a long-sequence modeling framework, named StreamPETR, for multi-view 3D object detection. Built upon the sparse query design in the PETR series, we systematically develop an object-centric temporal mechanism. The model is performed in an online manner and the long-term historical information is propagated through object queries frame by frame. Besides, we introduce a motion-aware layer normalization to model the movement of the objects. StreamPETR achieves significant performance improvements only with negligible computation cost, compared to the single-frame baseline. On the standard nuScenes benchmark, it is the first online multi-view method that achieves comparable performance (67.6% NDS & 65.3% AMOTA) with lidar-based methods. The lightweight version realizes 45.0% mAP and 31.7 FPS, outperforming the state-of-the-art method (SOLOFusion) by 2.3% mAP and 1.8× faster FPS. Code has been available at https://github.com/exiawsh/StreamPETR.git.

Referring video object segmentation (RVOS) aims at segmenting an object in a video following human instruction. Current state-of-the-art methods fall into an offline pattern, in which each clip independently interacts with text embedding for cross-modal understanding. They usually present that the offline pattern is necessary for RVOS, yet model limited temporal association within each clip. In this work, we break up the previous offline belief and propose a simple yet effective online model using explicit query propagation, named OnlineRefer. Specifically, our approach leverages target cues that gather semantic information and position prior to improve the accuracy and ease of referring predictions for the current frame. Furthermore, we generalize our online model into a semi-online framework to be compatible with video-based backbones. To show the effectiveness of our method, we evaluate it on four benchmarks, i.e., Refer-Youtube-VOS, Refer-DAVIS17, A2D-Sentences, and JHMDB-Sentences. Without bells and whistles, our OnlineRefer with a Swin-L backbone achieves 63.5 J&F and 64.8 J&F on Refer-Youtube-VOS and Refer-DAVIS17, outperforming all other offline methods. Our code is available at https://github.com/wudongming97/OnlineRefer.

In this paper, we propose a robust 3D detector, named Cross Modal Transformer (CMT), for end-to-end 3D multi-modal detection. Without explicit view transformation, CMT takes the image and point clouds tokens as inputs and directly outputs accurate 3D bounding boxes. The spatial alignment of multi-modal tokens is performed by encoding the 3D points into multi-modal features. The core design of CMT is quite simple while its performance is impressive. It achieves 74.1% NDS (state-of-the-art with single model) on nuScenes test set while maintaining faster inference speed. Moreover, CMT has a strong robustness even if the LiDAR is missing. Code is released at https://github.com/junjie18/CMT.

This paper proposes an efficient multi-camera to Bird’s-Eye-View (BEV) view transformation method for 3D perception, dubbed MatrixVT. Existing view transformers either suffer from poor efficiency or rely on device-specific operators, hindering the broad application of BEV models. In contrast, our method generates BEV features efficiently with only convolutions and matrix multiplications (MatMul). Specifically, we propose describing the BEV feature as the MatMul of image feature and a sparse Feature Transporting Matrix (FTM). A Prime Extraction module is then introduced to compress the dimension of image features and reduce FTM’s sparsity. Moreover, we propose the Ring & Ray Decomposition to replace the FTM with two matrices and reformulate our pipeline to reduce calculation further. Compared to existing methods, MatrixVT enjoys a faster speed and less memory footprint while remaining deploy-friendly. Extensive experiments on nuScenes and Waymo benchmarks demonstrate that our method is highly efficient but obtains results on par with the SOTA method in object detection and map segmentation tasks.

Mainstream 3D representation learning approaches are built upon contrastive or generative modeling pretext tasks, where great improvements in performance on various downstream tasks have been achieved. However, we find these two paradigms have different characteristics: (i) contrastive models are data-hungry that suffer from a representation over-fitting issue; (ii) generative models have a data filling issue that shows inferior data scaling capacity compared to contrastive models. This motivates us to learn 3D representations by sharing the merits of both paradigms, which is non-trivial due to the pattern difference between the two paradigms. In this paper, we propose contrast with reconstruct (ReCon) that unifies these two paradigms. ReCon is trained to learn from both generative modeling teachers and cross-modal contrastive teachers through ensemble distillation, where the generative student is used to guide the contrastive student. An encoder-decoder style ReCon-block is proposed that transfers knowledge through cross attention with stop-gradient, which avoids pretraining over-fitting and pattern difference issues. ReCon achieves a new state-of-the-art in 3D representation learning, e.g., 91.26% accuracy on ScanObjectNN. Codes have been released at https://github.com/qizekun/ReCon.

To enhance the reliability and robustness of language identification (LID) and language diarization (LD) systems for heterogeneous populations and scenarios, there is a need for speech processing models to be trained on datasets that feature diverse language registers and speech patterns. We present the MERLIon CCS challenge, featuring a first-of-its-kind Zoom video call dataset of parent-child shared book reading, of over 30 hours with over 300 recordings, annotated by multilingual transcribers using a high-fidelity linguistic transcription protocol. The audio corpus features spontaneous and in-the-wild English-Mandarin code-switching, child-directed speech in non-standard accents with diverse language-mixing patterns recorded in a variety of home environments. This report describes the corpus, as well as LID and LD results for our baseline and several systems submitted to the MERLIon CCS challenge using the corpus.

Data augmentation is a critical technique in object detection, especially the augmentations targeting at scale invariance training (scale-aware augmentation). However, there has been little systematic investigation of how to design scale-aware data augmentation for object detection. We propose Scale-aware AutoAug to learn data augmentation policies for object detection. We define a new scale-aware search space, where both image- and instance-level augmentations are designed for maintaining scale robust feature learning. Upon this search space, we propose a new search metric, termed Pareto Scale Balance, to facilitate efficient augmentation policy search. In experiments, Scale-aware AutoAug yields significant and consistent improvement on various object detectors (e.g., RetinaNet, Faster R-CNN, Mask R-CNN, and FCOS), even compared with strong multi-scale training baselines. Our searched augmentation policies are generalized well to other datasets and instance-level tasks beyond object detection, e.g., instance segmentation. The search cost is much less than previous automated augmentation approaches for object detection, i.e., 8 GPUs across 2.5 days versus. 800 TPU-days. In addition, meaningful patterns can be summarized from our searched policies, which intuitively provide valuable knowledge for hand-crafted data augmentation design. Based on the searched scale-aware augmentation policies, we further introduce a dynamic training paradigm to adaptively determine specific augmentation policy usage during training. The dynamic paradigm consists of an heuristic manner for image-level augmentations and a differentiable copy-paste-based method for instance-level augmentations. The dynamic paradigm achieves further performance improvements to Scale-aware AutoAug without any additional burden on the long tailed LVIS benchmarks. We also demonstrate its ability to prevent over-fitting for large models, e.g., the Swin Transformer large model. Code and models are available at https://github.com/dvlab-research/SA-AutoAug.

In this paper, we propose a novel query design for the transformer-based object detection. In previous transformer-based detectors, the object queries are a set of learned embeddings. However, each learned embedding does not have an explicit physical meaning and we cannot explain where it will focus on. It is difficult to optimize as the prediction slot of each object query does not have a specific mode. In other words, each object query will not focus on a specific region. To solve these problems, in our query design, object queries are based on anchor points, which are widely used in CNN-based detectors. So each object query focuses on the objects near the anchor point. Moreover, our query design can predict multiple objects at one position to solve the difficulty: ``one region, multiple objects''. In addition, we design an attention variant, which can reduce the memory cost while achieving similar or better performance than the standard attention in DETR. Thanks to the query design and the attention variant, the proposed detector that we called Anchor DETR, can achieve better performance and run faster than the DETR with 10x fewer training epochs. For example, it achieves 44.2 AP with 19 FPS on the MSCOCO dataset when using the ResNet50-DC5 feature for training 50 epochs. Extensive experiments on the MSCOCO benchmark prove the effectiveness of the proposed methods. Code is available at https://github.com/megvii-research/AnchorDETR.

In this paper, we propose the first self-distillation framework for general object detection, termed LGD (Label-Guided self-Distillation). Previous studies rely on a strong pretrained teacher to provide instructive knowledge that could be unavailable in real-world scenarios. Instead, we generate an instructive knowledge by inter-and-intra relation modeling among objects, requiring only student representations and regular labels. Concretely, our framework involves sparse label-appearance encoding, inter-object relation adaptation and intra-object knowledge mapping to obtain the instructive knowledge. They jointly form an implicit teacher at training phase, dynamically dependent on labels and evolving student representations. Modules in LGD are trained end-to-end with student detector and are discarded in inference. Experimentally, LGD obtains decent results on various detectors, datasets, and extensive tasks like instance segmentation. For example in MS-COCO dataset, LGD improves RetinaNet with ResNet-50 under 2x single-scale training from 36.2% to 39.0% mAP (+ 2.8%). It boosts much stronger detectors like FCOS with ResNeXt-101 DCN v2 under 2x multi-scale training from 46.1% to 47.9% (+ 1.8%). Compared with a classical teacher-based method FGFI, LGD not only performs better without requiring pretrained teacher but also reduces 51% training cost beyond inherent student learning.

Non-uniformed 3D sparse data, e.g., point clouds or voxels in different spatial positions, make contribution to the task of 3D object detection in different ways. Existing basic components in sparse convolutional networks (Sparse CNNs) process all sparse data, regardless of regular or submanifold sparse convolution. In this paper, we introduce two new modules to enhance the capability of Sparse CNNs, both are based on making feature sparsity learnable with position-wise importance prediction. They are focal sparse convolution (Focals Conv) and its multi-modal variant of focal sparse convolution with fusion, or Focals Conv-F for short. The new modules can readily substitute their plain counterparts in existing Sparse CNNs and be jointly trained in an end-to-end fashion. For the first time, we show that spatially learnable sparsity in sparse convolution is essential for sophisticated 3D object detection. Extensive experiments on the KITTI, nuScenes and Waymo benchmarks validate the effectiveness of our approach. Without bells and whistles, our results outperform all existing single-model entries on the nuScenes test benchmark. Code and models are at github.com/dvlab-research/FocalsConv.

We revisit large kernel design in modern convolutional neural networks (CNNs). Inspired by recent advances in vision transformers (ViTs), in this paper, we demonstrate that using a few large convolutional kernels instead of a stack of small kernels could be a more powerful paradigm. We suggested five guidelines, e.g., applying re-parameterized large depthwise convolutions, to design efficient high-performance large-kernel CNNs. Following the guidelines, we propose RepLKNet, a pure CNN architecture whose kernel size is as large as 31×31, in contrast to commonly used 3×3. RepLKNet greatly closes the performance gap between CNNs and ViTs, e.g., achieving comparable or superior results than Swin Transformer on ImageNet and a few typical downstream tasks, with lower latency. RepLKNet also shows nice scalability to big data and large models, obtaining 87.8% top-1 accuracy on ImageNet and 56.0% mIoU on ADE20K, which is very competitive among the state-of-the-arts with similar model sizes. Our study further reveals that, in contrast to small-kernel CNNs, large-kernel CNNs have much larger effective receptive fields and higher shape bias rather than texture bias. Code & models at https://github.com/megvii-research/RepLKNet.

Compared to convolutional layers, fully-connected (FC) layers are better at modeling the long-range dependencies but worse at capturing the local patterns, hence usually less favored for image recognition. In this paper, we propose a methodology, Locality Injection, to incorporate local priors into an FC layer via merging the trained parameters of a parallel conv kernel into the FC kernel. Locality Injection can be viewed as a novel Structural Re-parameterization method since it equivalently converts the structures via transforming the parameters. Based on that, we propose a multi-layer-perceptron (MLP) block named RepMLP Block, which uses three FC layers to extract features, and a novel architecture named RepMLPNet. The hierarchical design distinguishes RepMLPNet from the other concurrently proposed vision MLPs. As it produces feature maps of different levels, it qualifies as a backbone model for downstream tasks like semantic segmentation. Our results reveal that 1) Locality Injection is a general methodology for MLP models; 2) RepMLPNet has favorable accuracy-efficiency trade-off compared to the other MLPs; 3) RepMLPNet is the first MLP that seamlessly transfer to Cityscapes semantic segmentation. The code and models are available at https://github.com/DingXiaoH/RepMLP.

Long-tailed instance segmentation is a challenging task due to the extreme imbalance of training samples among classes. It causes severe biases of the head classes (with majority samples) against the tailed ones. This renders “how to appropriately define and alleviate the bias” one of the most important issues. Prior works mainly use label distribution or mean score information to indicate a coarse-grained bias. In this paper, we explore to excavate the confusion matrix, which carries the fine-grained misclassification details, to relieve the pairwise biases, generalizing the coarse one. To this end, we propose a novel Pairwise Class Balance (PCB) method, built upon a confusion matrix which is updated during training to accumulate the ongoing prediction preferences. PCB generates fightback soft labels for regularization during training. Besides, an iterative learning paradigm is developed to support a progressive and smooth regularization in such debiasing. PCB can be plugged and played to any existing method as a complement. Experimental results on LVIS demonstrate that our method achieves state-of-the-art performance without bells and whistles. Superior results across various architectures show the generalization ability. The code and trained models are available at https://github.com/megvii-research/PCB.

Sparsely annotated semantic segmentation (SASS) aims to train a segmentation network with coarse-grained (i.e., point-, scribble-, and block-wise) supervisions, where only a small proportion of pixels are labeled in each image. In this paper, we propose a novel tree energy loss for SASS by providing semantic guidance for unlabeled pixels. The tree energy loss represents images as minimum spanning trees to model both low-level and high-level pair-wise affini-ties. By sequentially applying these affinities to the net-work prediction, soft pseudo labels for unlabeled pixels are generated in a coarse-to-fine manner, achieving dynamic online self-training. The tree energy loss is effective and easy to be incorporated into existing frameworks by com-bining it with a traditional segmentation loss. Compared with previous SASS methods, our method requires no multi-stage training strategies, alternating optimization proce-dures, additional supervised data, or time-consuming post-processing while outperforming them in all SASS settings. Code is available at https://github.com/megvii-research/TreeEnergyLoss.

In this paper, we propose a new query-based detection framework for crowd detection. Previous query-based detectors suffer from two drawbacks: first, multiple predictions will be inferred for a single object, typically in crowded scenes; second, the performance saturates as the depth of the decoding stage increases. Benefiting from the nature of the one-to-one label assignment rule, we propose a progressive predicting method to address the above issues. Specifically, we first select accepted queries prone to generate true positive predictions, then refine the rest noisy queries according to the previously accepted predictions. Experiments show that our method can significantly boost the performance of query-based detectors in crowded scenes. Equipped with our approach, Sparse RCNN achieves 92.0% AP, 41.4% MR−2and 83.2% JI on the challenging CrowdHuman [35] dataset, outperforming the box-based method MIP [8] that specifies in handling crowded scenarios. Moreover, the proposed method, robust to crowdedness, can still obtain consistent improvements on moderately and slightly crowded datasets like CityPersons [47] and COCO [26]. Code will be made publicly available at https://github.com/megvii-model/Iter-E2EDET.

In this paper, we tackle the problem of learning visual representations from unlabeled scene-centric data. Existing works have demonstrated the potential of utilizing the underlying complex structure within scene-centric data; still, they commonly rely on hand-crafted objectness priors or specialized pretext tasks to build a learning framework, which may harm generalizability. Instead, we propose contrastive learning from data-driven semantic slots, namely SlotCon, for joint semantic grouping and representation learning. The semantic grouping is performed by assigning pixels to a set of learnable prototypes, which can adapt to each sample by attentive pooling over the feature and form new slots. Based on the learned data-dependent slots, a contrastive objective is employed for representation learning, which enhances the discriminability of features, and conversely facilitates grouping semantically coherent pixels together. Compared with previous efforts, by simultaneously optimizing the two coupled objectives of semantic grouping and contrastive learning, our approach bypasses the disadvantages of hand-crafted priors and is able to learn object/group-level representations from scene-centric images. Experiments show our approach effectively decomposes complex scenes into semantic groups for feature learning and significantly benefits downstream tasks, including object detection, instance segmentation, and semantic segmentation. Code is available at: https://github.com/CVMI-Lab/SlotCon.

In this paper, we explore the mask representation in instance segmentation with Point-of-Interest (PoI) features. Differentiating multiple potential instances within a single PoI feature is challenging, because learning a high-dimensional mask feature for each instance using vanilla convolution demands a heavy computing burden. To address this challenge, we propose an instance-aware convolution. It decomposes this mask representation learning task into two tractable modules as instance-aware weights and instance-agnostic features. The former is to parametrize convolution for producing mask features corresponding to different instances, improving mask learning efficiency by avoiding employing several independent convolutions. Meanwhile, the latter serves as mask templates in a single point. Together, instance-aware mask features are computed by convolving the template with dynamic weights, used for the mask prediction. Along with instance-aware convolution, we propose PointINS, a simple and practical instance segmentation approach, building upon dense one-stage detectors. Through extensive experiments, we evaluated the effectiveness of our framework built upon RetinaNet and FCOS. PointINS in ResNet101 backbone achieves a 38.3 mask mean average precision (mAP) on COCO dataset, outperforming existing point-based methods by a large margin. It gives a comparable performance to the region-based Mask R-CNN K. He, G. Gkioxari, P. Dollár, and R. Girshick, "Mask R-CNN," in Proc. IEEE Int. Conf. Comput. Vis., 2017, pp. 2980-2988 with faster inference.

In this paper, a simple yet effective network pruning framework is proposed to simultaneously address the problems of pruning indicator, pruning ratio, and efficiency constraint. This paper argues that the pruning decision should depend on the convolutional weights, and thus proposes novel weight-dependent gates (W-Gates) to learn the information from filter weights and obtain binary gates to prune or keep the filters automatically. To prune the network under efficiency constraints, a switchable Efficiency Module is constructed to predict the hardware latency or FLOPs of candidate pruned networks. Combined with the proposed Efficiency Module, W-Gates can perform filter pruning in an efficiency-aware manner and achieve a compact network with a better accuracy-efficiency trade-off. We have demonstrated the effectiveness of the proposed method on ResNet34, ResNet50, and MobileNet V2, respectively achieving up to 1.33/1.28/1.1 higher Top-1 accuracy with lower hardware latency on ImageNet. Compared with state-of-the-art methods, W-Gates also achieves superior performance.

Object detectors usually achieve promising results with the supervision of complete instance annotations. However, their performance is far from satisfactory with sparse instance annotations. Most existing methods for sparsely annotated object detection either re-weight the loss of hard negative samples or convert the unlabeled instances into ignored regions to reduce the interference of false negatives. We argue that these strategies are insufficient since they can at most alleviate the negative effect caused by missing annotations. In this paper, we propose a simple but effective mechanism, called Co-mining, for sparsely annotated object detection. In our Co-mining, two branches of a siamese network predict the pseudo-label sets for each other. To enhance multi-view learning and better mine unlabeled instances, the original image and corresponding augmented image are used as the inputs of two branches of the siamese network, respectively. Co-mining can serve as a general training mechanism applied to most of modern object detectors. Experiments are performed on MS COCO dataset with three different sparsely annotated settings using two typical frameworks: anchor-based detector RetinaNet and anchor-free detector FCOS. Experimental results show that our Co-mining with RetinaNet achieves 1.4%∼2.1% improvements compared with different baselines and surpasses existing methods under the same sparsely annotated setting.

We propose a novel point annotated setting for the weakly semi-supervised object detection task, in which the dataset comprises small fully annotated images and large weakly annotated images by points. It achieves a balance between tremendous annotation burden and detection performance. Based on this setting, we analyze existing detectors and find that these detectors have difficulty in fully exploiting the power of the annotated points. To solve this, we introduce a new detector, Point DETR, which extends DETR by adding a point encoder. Extensive experiments conducted on MS-COCO dataset in various data settings show the effectiveness of our method. In particular, when using 20% fully labeled data from COCO, our detector achieves a promising performance, 33.3 AP, which outperforms a strong baseline (FCOS) by 2.0 AP, and we demonstrate the point annotations bring over 10 points in various AR metrics.

We propose a new convolution called Dynamic Region-Aware Convolution (DRConv), which can automatically assign multiple filters to corresponding spatial regions where features have similar representation. In this way, DR-Conv outperforms standard convolution in modeling semantic variations. Standard convolutional layer can increase the number of filers to extract more visual elements but results in high computational cost. More gracefully, our DR-Conv transfers the increasing channel-wise filters to spatial dimension with learnable instructor, which not only improve representation ability of convolution, but also maintains computational cost and the translation-invariance as standard convolution dose. DRConv is an effective and elegant method for handling complex and variable spatial information distribution. It can substitute standard convolution in any existing networks for its plug-and-play property, especially to power convolution layers in efficient networks. We evaluate DRConv on a wide range of models (MobileNet series, ShuffleNetV2, etc.) and tasks (Classification, Face Recognition, Detection and Segmentation). On ImageNet classification, DRConv-based ShuffleNetV2-0.5 achieves state-of-the-art performance of 67.1% at 46M× multiply-adds level with 6.3% relative improvement.

This paper revisits feature pyramids networks (FPN) for one-stage detectors and points out that the success of FPN is due to its divide-and-conquer solution to the optimization problem in object detection rather than multi-scale feature fusion. From the perspective of optimization, we introduce an alternative way to address the problem instead of adopting the complex feature pyramids - utilizing only one-level feature for detection. Based on the simple and efficient solution, we present You Only Look One-level Feature (YOLOF). In our method, two key components, Dilated Encoder and Uniform Matching, are proposed and bring considerable improvements. Extensive experiments on the COCO benchmark prove the effectiveness of the proposed model. Our YOLOF achieves comparable results with its feature pyramids counterpart RetinaNet while being 2.5× faster. Without transformer layers, YOLOF can match the performance of DETR in a single-level feature manner with 7× less training epochs. Code is available at https://github.com/megvii-model/YOLOF.

We propose a universal building block of Convolutional Neural Network (ConvNet) to improve the performance without any inference-time costs. The block is named Diverse Branch Block (DBB), which enhances the representational capacity of a single convolution by combining diverse branches of different scales and complexities to enrich the feature space, including sequences of convolutions, multiscale convolutions, and average pooling. After training, a DBB can be equivalently converted into a single conv layer for deployment. Unlike the advancements of novel ConvNet architectures, DBB complicates the training-time microstructure while maintaining the macro architecture, so that it can be used as a drop-in replacement for regular conv layers of any architecture. In this way, the model can be trained to reach a higher level of performance and then transformed into the original inference-time structure for inference. DBB improves ConvNets on image classification (up to 1.9% higher top-1 accuracy on ImageNet), object detection and semantic segmentation. The PyTorch code and models are released at https://github.com/DingXiaoH/DiverseBranchBlock.

We present a simple but powerful architecture of convolutional neural network, which has a VGG-like inference-time body composed of nothing but a stack of 3 × 3 convolution and ReLU, while the training-time model has a multi-branch topology. Such decoupling of the training-time and inference-time architecture is realized by a structural re-parameterization technique so that the model is named RepVGG. On ImageNet, RepVGG reaches over 80% top-1 accuracy, which is the first time for a plain model, to the best of our knowledge. On NVIDIA 1080Ti GPU, RepVGG models run 83% faster than ResNet-50 or 101% faster than ResNet-101 with higher accuracy and show favorable accuracy-speed trade-off compared to the state-of-the-art models like EfficientNet and RegNet. The code and trained models are available at https://github.com/megvii-model/RepVGG.

We present a simple, effective, and general activation function we term ACON which learns to activate the neurons or not. Interestingly, we find Swish, the recent popular NAS-searched activation, can be interpreted as a smooth approximation to ReLU. Intuitively, in the same way, we approximate the more general Maxout family to our novel ACON family, which remarkably improves the performance and makes Swish a special case of ACON. Next, we present meta-ACON, which explicitly learns to optimize the parameter switching between non-linear (activate) and linear (inactivate) and provides a new design space. By simply changing the activation function, we show its effectiveness on both small models and highly optimized large models (e.g. it improves the ImageNet top-1 accuracy rate by 6.7% and 1.8% on MobileNet-0.25 and ResNet-152, respectively). Moreover, our novel ACON can be naturally transferred to object detection and semantic segmentation, showing that ACON is an effective alternative in a variety of tasks. Code is available at https://github.com/nmaac/acon.

In this paper, we investigate a new variant of neural architecture search (NAS) paradigm – searching with random labels (RLNAS). The task sounds counter-intuitive for most existing NAS algorithms since random label provides few information on the performance of each candidate architecture. Instead, we propose a novel NAS framework based on ease-of-convergence hypothesis, which requires only random labels during searching. The algorithm involves two steps: first, we train a SuperNet using random labels; second, from the SuperNet we extract the subnetwork whose weights change most significantly during the training. Extensive experiments are evaluated on multiple datasets (e.g. NAS-Bench-201 and ImageNet) and multiple search spaces (e.g. DARTS-like and MobileNet-like). Very surprisingly, RLNAS achieves comparable or even better results compared with state-of-the-art NAS methods such as PC-DARTS, Single Path One-Shot, even though the counterparts utilize full ground truth labels for searching. We hope our finding could inspire new understandings on the essential of NAS.

In this paper, we present a novel approach to synthesize realistic images based on their semantic layouts. It hypothesizes that for objects with similar appearance, they share similar representation. Our method establishes dependencies between regions according to their appearance correlation, yielding both spatially variant and associated representations. Conditioning on these features, we propose a dynamic weighted network constructed by spatially conditional computation (with both convolution and normalization). More than preserving semantic distinctions, the given dynamic network strengthens semantic relevance, benefiting global structure and detail synthesis. We demonstrate that our method gives the compelling generation performance qualitatively and quantitatively with extensive experiments on benchmarks.

We propose a novel implicit feature refinement module for high-quality instance segmentation. Existing image/video instance segmentation methods rely on explicitly stacked convolutions to refine instance features before the final prediction. In this paper, we first give an empirical comparison of different refinement strategies, which reveals that the widely-used four consecutive convolutions are not necessary. As an alternative, weight-sharing convolution blocks provides competitive performance. When such block is iterated for infinite times, the block output will eventually converge to an equilibrium state. Based on this observation, the implicit feature refinement (IFR) is developed by constructing an implicit function. The equilibrium state of instance features can be obtained by fixed-point iteration via a simulated infinite-depth network. Our IFR enjoys several advantages: 1) simulates an infinite-depth refinement network while only requiring parameters of single residual block; 2) produces high-level equilibrium instance features of global receptive field; 3) serves as a plug-and-play general module easily extended to most object recognition frameworks. Experiments on the COCO and YouTube-VIS benchmarks show that our IFR achieves improved performance on state-of-the-art image/video instance segmentation frameworks, while reducing the parameter burden (e.g. 1% AP improvement on Mask R-CNN with only 30.0% parameters in mask head). Code will be made available at \hrefhttps://github.com/lufanma/IFR.git https://github.com/lufanma/IFR.git .

In this paper, we propose an end-to-end framework for instance segmentation. Based on the recently introduced DETR, our method, termed SOLQ, segments objects by learning unified queries. In SOLQ, each query represents one object and has multiple representations: class, location and mask. The object queries learned perform classification, box regression and mask encoding simultaneously in an unified vector form. During training phase, the mask vectors encoded are supervised by the compression coding of raw spatial masks. In inference time,mask vectors produced can be directly transformed to spatial masks by the inverse process of compression coding. Experimental results show that SOLQ can achieve state-of-the-art performance, surpassing most of existing approaches. Moreover, the joint learning of unified query representation can greatly improve the detection performance of DETR. We hope our SOLQ can serve as a strong baseline for the Transformer-based instance segmentation.

Knowledge distillation has shown great success in classification, however, it is still challenging for detection. In a typical image for detection, representations from different locations may have different contributions to detection targets, making the distillation hard to balance. In this paper, we propose a conditional distillation framework to distill the desired knowledge, namely knowledge that is beneficial in terms of both classification and localization for every instance. The framework introduces a learnable conditional decoding module, which retrieves information given each target instance as query. Specifically, we encode the condition information as query and use the teacher's representations as key. The attention between query and key is used to measure the contribution of different features, guided by a localization-recognition-sensitive auxiliary task. Extensive experiments demonstrate the efficacy of our method: we observe impressive improvements under various settings. Notably, we boost RetinaNet with ResNet-50 backbone from $37.4$ to $40.7$ mAP ($+3.3$) under $1\times$ schedule, that even surpasses the teacher ($40.4$ mAP) with ResNet-101 backbone under $3\times$ schedule. Code has been released on https://github.com/megvii-research/ICD.

In this paper, we comprehensively reveal the learning dynamics of normalized neural network using Stochastic Gradient Descent (with momentum) and Weight Decay (WD), named as Spherical Motion Dynamics (SMD). Most related works focus on studying behavior of effective learning rate" inequilibrium" state, i.e. assuming weight norm remains unchanged. However, their discussion on why this equilibrium can be reached is either absent or less convincing. Our work directly explores the cause of equilibrium, as a special state of SMD. Specifically, 1) we introduce the assumptions that can lead to equilibrium state in SMD, and prove equilibrium can be reached in a linear rate regime under given assumptions; 2) we propose ``angular update" as a substitute for effective learning rate to depict the state of SMD, and derive the theoretical value of angular update in equilibrium state; 3) we verify our assumptions and theoretical results on various large-scale computer vision tasks including ImageNet and MSCOCO with standard settings. Experiment results show our theoretical findings agree well with empirical observations. We also show that the behavior of angular update in SMD can produce interesting effect to the optimization of neural network in practice.

We present joint multi-dimension pruning (abbreviated as JointPruning), an effective method of pruning a network on three crucial aspects: spatial, depth and channel simultaneously. To tackle these three naturally different dimensions, we proposed a general framework by defining pruning as seeking the best pruning vector (i.e., the numerical value of layer-wise channel number, spatial size, depth) and construct a unique mapping from the pruning vector to the pruned network structures. Then we optimize the pruning vector with gradient update and model joint pruning as a numerical gradient optimization process. To overcome the challenge that there is no explicit function between the loss and the pruning vectors, we proposed self-adapted stochastic gradient estimation to construct a gradient path through network loss to pruning vectors and enable efficient gradient update. We show that the joint strategy discovers a better status than previous studies that focused on individual dimensions solely, as our method is optimized collaboratively across the three dimensions in a single end-to-end training and it is more efficient than the previous exhaustive methods. Extensive experiments on large-scale ImageNet dataset across a variety of network architectures MobileNet V1&V2&V3 and ResNet demonstrate the effectiveness of our proposed method. For instance, we achieve significant margins of 2.5% and 2.6% improvement over the state-of-the-art approach on the already compact MobileNet V1&V2 under an extremely large compression ratio.

We propose a simple yet effective proposal-based object detector, aiming at detecting highly-overlapped instances in crowded scenes. The key of our approach is to let each proposal predict a set of correlated instances rather than a single one in previous proposal-based frameworks. Equipped with new techniques such as EMD Loss and Set NMS, our detector can effectively handle the difficulty of detecting highly overlapped objects. On a FPN-Res50 baseline, our detector can obtain 4.9\% AP gains on challenging CrowdHuman dataset and 1.0\% $\text{MR}^{-2}$ improvements on CityPersons dataset, without bells and whistles. Moreover, on less crowed datasets like COCO, our approach can still achieve moderate improvement, suggesting the proposed method is robust to crowdedness.

Recently, numerous handcrafted and searched networks have been applied for semantic segmentation. However, previous works intend to handle inputs with various scales in pre-defined static architectures, such as FCN, U-Net, and DeepLab series. This paper studies a conceptually new method to alleviate the scale variance in semantic representation, named dynamic routing. The proposed framework generates data-dependent routes, adapting to the scale distribution of each image. To this end, a differentiable gating function, called soft conditional gate, is proposed to select scale transform paths on the fly. In addition, the computational cost can be further reduced in an end-to-end manner by giving budget constraints to the gating function. We further relax the network level routing space to support multi-path propagations and skip-connections in each forward, bringing substantial network capacity. To demonstrate the superiority of the dynamic property, we compare with several static architectures, which can be modeled as special cases in the routing space. Extensive experiments are conducted on Cityscapes and PASCAL VOC 2012 to illustrate the effectiveness of the dynamic framework. Code is available at https://github.com/yanwei-li/DynamicRouting.

Traditional convolution-based generative adversarial networks synthesize images based on hierarchical local operations, where long-range dependency relation is implicitly modeled with a Markov chain. It is still not sufficient for categories with complicated structures. In this paper, we characterize long-range dependence with attentive normalization (AN), which is an extension to traditional instance normalization. Specifically, the input feature map is softly divided into several regions based on its internal semantic similarity, which are respectively normalized. It enhances consistency between distant regions with semantic correspondence. Compared with self-attention GAN, our attentive normalization does not need to measure the correlation of all locations, and thus can be directly applied to large-size feature maps without much computational burden. Extensive experiments on class-conditional image generation and semantic inpainting verify the efficacy of our proposed module.

Understanding interactions between humans and objects is one of the fundamental problems in visual classification and an essential step towards detailed scene understanding. Human-object interaction (HOI) detection strives to localize both the human and an object as well as the identification of complex interactions between them. Most existing HOI detection approaches are instance-centric where interactions between all possible human-object pairs are predicted based on appearance features and coarse spatial information. We argue that appearance features alone are insufficient to capture complex human-object interactions. In this paper, we therefore propose a novel fully-convolutional approach that directly detects the interactions between human-object pairs. Our network predicts interaction points, which directly localize and classify the inter-action. Paired with the densely predicted interaction vectors, the interactions are associated with human and object detections to obtain final predictions. To the best of our knowledge, we are the first to propose an approach where HOI detection is posed as a keypoint detection and grouping problem. Experiments are performed on two popular benchmarks: V-COCO and HICO-DET. Our approach sets a new state-of-the-art on both datasets. Code is available at https://github.com/vaesl/IP-Net.

The Learnable Tree Filter presents a remarkable approach to model structure-preserving relations for semantic segmentation. Nevertheless, the intrinsic geometric constraint forces it to focus on the regions with close spatial distance, hindering the effective long-range interactions. To relax the geometric constraint, we give the analysis by reformulating it as a Markov Random Field and introduce a learnable unary term. Besides, we propose a learnable spanning tree algorithm to replace the original non-differentiable one, which further improves the flexibility and robustness. With the above improvements, our method can better capture long range dependencies and preserve structural details with linear complexity, which is extended to several vision tasks for more generic feature transform. Extensive experiments on object detection/instance segmentation demonstrate the consistent improvements over the original version. For semantic segmentation, we achieve leading performance (82.1% mIoU) on the Cityscapes benchmark without bells-and whistles. Code is available at https://github.com/StevenGrove/LearnableTreeFilterV2.

Large-scale object detection datasets (e.g., MS-COCO) try to define the ground truth bounding boxes as clear as possible. However, we observe that ambiguities are still introduced when labeling the bounding boxes. In this paper, we propose a novel bounding box regression loss for learning bounding box transformation and localization variance together. Our loss greatly improves the localization accuracies of various architectures with nearly no additional computation. The learned localization variance allows us to merge neighboring bounding boxes during non-maximum suppression (NMS), which further improves the localization performance. On MS-COCO, we boost the Average Precision (AP) of VGG-16 Faster R-CNN from 23.6% to 29.1%. More importantly, for ResNet-50-FPN Mask R-CNN, our method improves the AP and AP90 by 1.8% and 6.2% respectively, which significantly outperforms previous state-of-the-art bounding box refinement methods. Our code and models are available at github.com/yihui-he/KL-Loss.

Recent research on super-resolution has achieved great success due to the development of deep convolutional neural networks (DCNNs). However, super-resolution of arbitrary scale factor has been ignored for a long time. Most previous researchers regard super-resolution of differentscale factors as independent tasks. They train a specific model for each scale factor which is inefficient in computing, and prior work only take the super-resolution of several integer scale factors into consideration. In this work,we propose a novel method called Meta-SR to firstly solve super-resolution of arbitrary scale factor (including non-integer scale factors) with a single model. In our Meta-SR,the Meta-Upscale Module is proposed to replace the traditional upscale module. For arbitrary scale factor, the Meta-Upscale Module dynamically predicts the weights of the up-scale filters by taking the scale factor as input and use these weights to generate the HR image of arbitrary size. For any low-resolution image, our Meta-SR can continuously zoomin it with arbitrary scale factor by only using a single model.We evaluated the proposed method through extensive experiments on widely used benchmark datasets on single image super-resolution. The experimental results show the superiority of our Meta-Upscale.

In this paper, we introduce a new large-scale object detection dataset, Objects365, which has 365 object categories over 600K training images. More than 10 million, high-quality bounding boxes are manually labeled through a three-step, carefully designed annotation pipeline. It is the largest object detection dataset (with full annotation) so far and establishes a more challenging benchmark for the community. Objects365 can serve as a better feature learning dataset for localization-sensitive tasks like object detection and semantic segmentation. The Objects365 pre-trained models significantly outperform ImageNet pre-trained models with 5.6 points gain (42 vs 36.4) based on the standard setting of 90K iterations on COCO benchmark. Even compared with much long training time like 540K iterations, our Objects365 pretrained model with 90K iterations still have 2.7 points gain (42 vs 39.3). Meanwhile, the finetuning time can be greatly reduced (up to 10 times) when reaching the same accuracy. Better generalization ability of Object365 has also been verified on CityPersons, VOC segmentation, and ADE tasks. The dataset as well as the pretrained-models have been released at www.objects365.org.

In this paper, we propose a novel meta learning approach for automatic channel pruning of very deep neural networks. We first train a PruningNet, a kind of meta network, which is able to generate weight parameters for any pruned structure given the target network. We use a simple stochastic structure sampling method for training the PruningNet. Then, we apply an evolutionary procedure to search for good-performing pruned networks. The search is highly efficient because the weights are directly generated by the trained PruningNet and we do not need any finetuning at search time. With a single PruningNet trained for the target network, we can search for various Pruned Networks under different constraints with little human participation. Compared to the state-of-the-art pruning methods, we have demonstrated superior performances on MobileNet V1/V2 and ResNet. Codes are available on https://github.com/liuzechun/MetaPruning.

Object detectors are usually equipped with backbone networks designed for image classification. It might be sub-optimal because of the gap between the tasks of image classification and object detection. In this work, we present DetNAS to use Neural Architecture Search (NAS) for the design of better backbones for object detection. It is non-trivial because detection training typically needs ImageNetpre-training while NAS systems require accuracies on the target detection task as supervisory signals. Based on the technique of one-shot supernet, which contains all possible networks in the search space, we propose a framework for backbone search on object detection. We train the supernet under the typical detector training schedule: ImageNet pre-training and detection fine-tuning. Then, the architecture search is performed on the trained supernet, using the detection task as the guidance. This framework makes NAS on backbones very efficient. In experiments, we show the effectiveness of DetNAS on various detectors, for instance, one-stage RetinaNetand the two-stage FPN. We empirically find that networks searched on object detection shows consistent superiority compared to those searched on ImageNet classification. The resulting architecture achieves superior performance than hand-crafted networks on COCO with much less FLOPs complexity.

The development of object detection in the era of deep learning, from R-CNN [11], Fast/Faster R-CNN [10, 31] to recent Mask R-CNN [14] and RetinaNet [24], mainly come from novel network, new framework, or loss design. However, mini-batch size, a key factor for the training of deep neural networks, has not been well studied for object detection. In this paper, we propose a Large Mini-Batch Object Detector (MegDet) to enable the training with a large mini-batch size up to 256, so that we can effectively utilize at most 128 GPUs to significantly shorten the training time. Technically, we suggest a warmup learning rate policy and Cross-GPU Batch Normalization, which together allow us to successfully train a large mini-batch detector in much less time (e.g., from 33 hours to 4 hours), and achieve even better accuracy. The MegDet is the backbone of our submission (mmAP 52.5%) to COCO 2017 Challenge, where we won the 1st place of Detection task.

We introduce an extremely computation-efficient CNN architecture named ShuffleNet, which is designed specially for mobile devices with very limited computing power (e.g., 10-150 MFLOPs). The new architecture utilizes two new operations, pointwise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy. Experiments on ImageNet classification and MS COCO object detection demonstrate the superior performance of ShuffleNet over other structures, e.g. lower top-1 error (absolute 7.8%) than recent MobileNet [12] on ImageNet classification task, under the computation budget of 40 MFLOPs. On an ARM-based mobile device, ShuffleNet achieves ~13× actual speedup over AlexNet while maintaining comparable accuracy.

We propose a novel and flexible anchor mechanism named MetaAnchor for object detection frameworks. Unlike many previous detectors model anchors via a predefined manner, in MetaAnchor anchor functions could be dynamically generated from the arbitrary customized prior boxes. Taking advantage of weight prediction, MetaAnchor is able to work with most of the anchor-based object detection systems such as RetinaNet. Compared with the predefined anchor scheme, we empirically find that MetaAnchor is more robust to anchor settings and bounding box distributions; in addition, it also shows the potential on the transfer task. Our experiment on COCO detection task shows MetaAnchor consistently outperforms the counterparts in various scenarios.

One of recent trends [31, 32, 14] in network architecture design is stacking small filters (e.g., 1×1 or 3×3) in the entire network because the stacked small filters is more efficient than a large kernel, given the same computational complexity. However, in the field of semantic segmentation, where we need to perform dense per-pixel prediction, we find that the large kernel (and effective receptive field) plays an important role when we have to perform the classification and localization tasks simultaneously. Following our design principle, we propose a Global Convolutional Network to address both the classification and localization issues for the semantic segmentation. We also suggest a residual-based boundary refinement to further refine the object boundaries. Our approach achieves state-of-art performance on two public benchmarks and significantly outperforms previous results, 82.2% (vs 80.2%) on PASCAL VOC 2012 dataset and 76.9% (vs 71.8%) on Cityscapes dataset.

In this paper, we introduce a new channel pruning method to accelerate very deep convolutional neural networks. Given a trained CNN model, we propose an iterative two-step algorithm to effectively prune each layer, by a LASSO regression based channel selection and least square reconstruction. We further generalize this algorithm to multi-layer and multi-branch cases. Our method reduces the accumulated error and enhance the compatibility with various architectures. Our pruned VGG-16 achieves the state-of-the-art results by 5× speed-up along with only 0.3% increase of error. More importantly, our method is able to accelerate modern networks like ResNet, Xception and suffers only 1.4%, 1.0% accuracy loss under 2× speedup respectively, which is significant.

Most object detectors contain two important components: a feature extractor and an object classifier. The feature extractor has rapidly evolved with significant research efforts leading to better deep convolutional architectures. The object classifier, however, has not received much attention and many recent systems (like SPPnet and Fast/Faster R-CNN) use simple multi-layer perceptrons. This paper demonstrates that carefully designing deep networks for object classification is just as important. We experiment with region-wise classifier networks that use shared, region-independent convolutional features. We call them "Networks on Convolutional feature maps" (NoCs). We discover that aside from deep feature maps, a deep and convolutional per-region classifier is of particular importance for object detection, whereas latest superior image classification models (such as ResNets and GoogLeNets) do not directly lead to good detection accuracy without using such a per-region classifier. We show by experiments that despite the effective ResNets and Faster R-CNN systems, the design of NoCs is an essential element for the 1st-place winning entries in ImageNet and MS COCO challenges 2015.

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers - 8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

This paper aims to accelerate the test-time computation of convolutional neural networks (CNNs), especially very deep CNNs [1] that have substantially impacted the computer vision community. Unlike previous methods that are designed for approximating linear filters or linear responses, our method takes the nonlinear units into account. We develop an effective solution to the resulting nonlinear optimization problem without the need of stochastic gradient descent (SGD). More importantly, while previous methods mainly focus on optimizing one or two layers, our nonlinear method enables an asymmetric reconstruction that reduces the rapidly accumulated error when multiple (e.g., ≥ 10) layers are approximated. For the widely used very deep VGG-16 model [1] , our method achieves a whole-model speedup of 4 × with merely a 0.3 percent increase of top-5 error in ImageNet classification. Our 4 × accelerated VGG-16 model also shows a graceful accuracy degradation for object detection when plugged into the Fast R-CNN detector [2] .

This paper aims to accelerate the test-time computation of deep convolutional neural networks (CNNs). Unlike existing methods that are designed for approximating linear filters or linear responses, our method takes the nonlinear units into account. We minimize the reconstruction error of the nonlinear responses, subject to a low-rank constraint which helps to reduce the complexity of filters. We develop an effective solution to this constrained nonlinear optimization problem. An algorithm is also presented for reducing the accumulated error when multiple layers are approximated. A whole-model speedup ratio of 4× is demonstrated on a large network trained for ImageNet, while the top-5 error rate is only increased by 0.9%. Our accelerated model has a comparably fast speed as the “AlexNet” [11], but is 4.7% more accurate.

Rectified activation units (rectifiers) are essential for state-of-the-art neural networks. In this work, we study rectifier neural networks for image classification from two aspects. First, we propose a Parametric Rectified Linear Unit (PReLU) that generalizes the traditional rectified unit. PReLU improves model fitting with nearly zero extra computational cost and little overfitting risk. Second, we derive a robust initialization method that particularly considers the rectifier nonlinearities. This method enables us to train extremely deep rectified models directly from scratch and to investigate deeper or wider network architectures. Based on the learnable activation and advanced initialization, we achieve 4.94% top-5 test error on the ImageNet 2012 classification dataset. This is a 26% relative improvement over the ILSVRC 2014 winner (GoogLeNet, 6.66% [33]). To our knowledge, our result is the first to surpass the reported human-level performance (5.1%, [26]) on this dataset.

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224 × 224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102 × faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

In the light of the practical project experience, the multi-terminal DC (MTDC) is regarded as one of the preferable solutions to solve the grid interconnection issue of wind generation. This paper mainly focuses on the application of the voltage source converter(VSC) based MTDC technology to integrate large scale wind farms to the electric power grid. A radial MTDC system is explored as the best choice for wind power integration, due to that it can mitigate the fluctuation of the aggregated wind power. Based on the analysis of the VSC model and control, the coordinated control strategy for the proposed MTDC system is designed. The operation performance of a four-terminal MTDC system connecting two DFIG-based wind farms, the local and remote grids is also studied, and the proposed control strategy is adopted to achieve a constant power for long distant transmission to the load center under wind speed variations and faults on DC line.