Recent vision-language-action (VLA) models built on pretrained vision-language models (VLMs) have demonstrated strong performance in robotic manipulation. However, these models remain constrained by the single-frame image paradigm and fail to fully leverage the temporal information offered by multi-frame histories, as directly feeding multiple frames into VLM backbones incurs substantial computational overhead and inference latency. We propose CronusVLA, a unified framework that extends single-frame VLA models to the multi-frame paradigm. CronusVLA follows a two-stage process: (1) Single-frame pretraining on large-scale embodied datasets with autoregressive prediction of action tokens, establishing an effective embodied vision-language foundation; (2) Multi-frame post-training, which adapts the prediction of the vision-language backbone from discrete tokens to learnable features, and aggregates historical information via feature chunking. CronusVLA effectively addresses the existing challenges of multi-frame modeling while enhancing performance. To evaluate the robustness under temporal and spatial disturbances, we introduce SimplerEnv-OR, a novel benchmark featuring 24 types of observational disturbances and 120 severity levels. Experiments across three embodiments in simulated and real-world environments demonstrate that CronusVLA achieves leading performance and superior robustness, with a 70.9% success rate on SimplerEnv, a 26.8% improvement over OpenVLA on LIBERO, and the highest robustness score on SimplerEnv-OR, showing the promise of efficient multi-frame adaptation for real-world VLA deployment.

Large language models have achieved substantial progress in mathematical reasoning, yet their advancement is limited by the scarcity of high-quality, high-difficulty training data. Existing synthesis methods largely rely on transforming human-written templates, limiting both diversity and scalability. We propose MathSmith, a novel framework for synthesizing challenging mathematical problems to enhance LLM reasoning. Rather than modifying existing problems, MathSmith constructs new ones from scratch by randomly sampling concept–explanation pairs from PlanetMath, ensuring data independence and avoiding contamination. To increase difficulty, we design nine predefined strategies as soft constraints during rationales. We further adopts reinforcement learning to jointly optimize structural validity, reasoning complexity, and answer consistency. The length of the reasoning trace generated under autoregressive prompting is used to reflect cognitive complexity, encouraging the creation of more demanding problems aligned with long-chain-of-thought reasoning. Experiments across five benchmarks, categorized as easy & medium (GSM8K, MATH-500) and hard (AIME2024, AIME2025, OlympiadBench), show that MathSmith consistently outperforms existing baselines under both short and long CoT settings. Additionally, a weakness-focused variant generation module enables targeted improvement on specific concepts. Overall, MathSmith exhibits strong scalability, generalization, and transferability, highlighting the promise of high-difficulty synthetic data in advancing LLM reasoning capabilities.

Yu Li, Xiaoran Shang, Qizhi Pei, Yun Zhu, Xin Gao, Honglin Lin, Zhanping Zhong, Zhuoshi Pan, Zheng Liu, Xiaoyang Wang, Conghui He, Dahua Lin, Feng Zhao, Lijun Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Yichuan Ma, Linyang Li, Yongkang Chen, Peiji Li, Xiaozhe Li, Qipeng Guo, Dahua Lin, Kai Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Hybrid parallelism underpins large-scale LLM training across tens of thousands of GPUs. At such scale, hardware failures on individual devices lead to performance skew across devices, diminishing overall training efficiency. Existing resilient systems overlook sequence length variability in datasets and device performance skew under hybrid parallelism. As a result, (1) iteration time fluctuations induced by sequence length variability can trigger spurious fail-slow detections, and (2) failures are mitigated through individual adaptations in hybrid parallelism, leading to unnecessary detection overhead and inefficient resilient training.

Video generation has witnessed significant advancements, yet evaluating these models remains a challenge. A comprehensive evaluation benchmark for video generation is indispensable for two reasons: 1) Existing metrics do not fully align with human perceptions; 2) An ideal evaluation system should provide insights to inform future developments of video generation. To this end, we present VBench++, a comprehensive benchmark suite that dissects "video generation quality" into specific, hierarchical, and disentangled dimensions, each with tailored prompts and evaluation methods. VBench++ has several appealing properties: 1) Comprehensive Dimensions: VBench++ comprises 16 dimensions in text-to-video generation (e.g., subject identity inconsistency, motion smoothness, temporal flickering, and spatial relationship, etc). The evaluation metrics with fine-grained levels reveal individual models' strengths and weaknesses. 2) Human Alignment: We also provide a dataset of human preference annotations to validate our benchmarks' alignment with human perception, for each evaluation dimension respectively. 3) Valuable Insights: We look into current models' ability across various evaluation dimensions, and various content types. We also investigate the gaps between video and image generation models. 4) Versatile Benchmarking: VBench++ is designed to evaluate a wide range of video generation tasks, including text-to-video and image-to-video. We introduce a high-quality Image Suite with an adaptive aspect ratio to enable fair evaluations across different image-to-video generation settings. Beyond assessing technical quality, VBench++ evaluates the trustworthiness of video generative models, providing a more holistic view of model performance. 5) Full Open-Sourcing: We fully open-source VBench++, including all prompts, the Image Suite, evaluation methods, generated videos, and human preference annotations.

CLIP (Contrastive Language-Image Pre-training) uses contrastive learning from noise image-text pairs to excel at recognizing a wide array of candidates, yet its focus on broad associations hinders the precision in distinguishing subtle differences among fine-grained items. Conversely, Multimodal Large Language Models (MLLMs) excel at classifying fine-grained categories, thanks to their substantial knowledge from pre-training on web-level corpora. However, the performance of MLLMs declines with an increase in category numbers, primarily due to growing complexity and constraints of limited context window size. To synergize the strengths of both approaches and enhance the few-shot/zero-shot recognition abilities for datasets characterized by extensive and fine-grained vocabularies, this paper introduces RAR, a Retrieving And Ranking augmented method for MLLMs. We initially establish a multi-modal retriever based on CLIP to create and store explicit memory for different categories beyond the immediate context window. During inference, RAR retrieves the top- $k$ similar results from the memory and uses MLLMs to rank and make the final predictions. Our proposed approach not only addresses the inherent limitations in fine-grained recognition but also preserves the model's comprehensive knowledge base, significantly boosting accuracy across a range of vision-language recognition tasks. Notably, our approach demonstrates a significant improvement in performance on 5 fine-grained visual recognition benchmarks, 11 few-shot image recognition datasets, and the 2 object detection datasets under the zero-shot recognition setting.

Recent evaluations of Large Multimodal Models (LMMs) have explored their capabilities in various domains, with only few benchmarks specifically focusing on urban environments. Moreover, existing urban benchmarks have been limited to evaluating LMMs with basic region-level urban tasks under singular views, leading to incomplete evaluations of LMMs' abilities in urban environments. To address these issues, we present UrBench, a comprehensive benchmark designed for evaluating LMMs in complex multi-view urban scenarios. UrBench contains 11.6K meticulously curated questions at both region-level and role-level that cover 4 task dimensions: Geo-Localization, Scene Reasoning, Scene Understanding, and Object Understanding, totaling 14 task types. In constructing UrBench, we utilize data from existing datasets and additionally collect data from 11 cities, creating new annotations using a cross-view detection-matching method. With these images and annotations, we then integrate LMM-based, rule-based, and human-based methods to construct large-scale high-quality questions. Our evaluations on 21 LMMs show that current LMMs struggle in the urban environments in several aspects. Even the best performing GPT-4o lags behind humans in most tasks, ranging from simple tasks such as counting to complex tasks such as orientation, localization and object attribute recognition, with an average performance gap of 17.4%. Our benchmark also reveals that LMMs exhibit inconsistent behaviors with different urban views, especially with respect to understanding cross-view relations.

Refusal-Aware Instruction Tuning (RAIT) enables Large Language Models (LLMs) to refuse to answer unknown questions. By modifying responses of unknown questions in the training data to refusal responses such as ''I don't know", RAIT enhances the reliability of LLMs and reduces their hallucination. Generally, RAIT modifies training samples based on the correctness of the initial LLM's response. However, this crude approach can cause LLMs to excessively refuse answering questions they could have correctly answered, the problem we call over-refusal. In this paper, we explore two primary causes of over-refusal: Static conflict occurs when similar samples within the LLM’s feature space receive differing supervision signals (original vs. modified ''I don't know"). Dynamic conflict arises as the LLM's evolving knowledge during SFT enables it to answer previously unanswerable questions, but the now-answerable training samples still retain the original ''I don't know" supervision signals from the initial LLM state, leading to inconsistencies. These conflicts cause the trained LLM to misclassify known questions as unknown, resulting in over-refusal. To address this issue, we introduce Certainty Represented Knowledge Flow for Refusal-Aware Instructions Tuning (CRaFT). CRaFT centers on two main contributions: First, we additionally incorporate response certainty to selectively filter and modify data, reducing static conflicts. Second, we implement preliminary rehearsal training to characterize changes in the LLM's knowledge state, which helps mitigate dynamic conflicts during the fine-tuning process. We conducted extensive experiments on open-ended question answering and multiple-choice question task. Experiment results show that CRaFT can improve LLM's overall performance during the RAIT process.

Zhi Chen, Qiguang Chen, Libo Qin, Qipeng Guo, Haijun Lv, Yicheng Zou, Hang Yan, Kai Chen, Dahua Lin. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Creating lyrics and melodies for the vocal track in a symbolic format, known as song composition, demands expert musical knowledge of melody, an advanced understanding of lyrics, and precise alignment between them. Despite achievements in sub-tasks such as lyric generation, lyric-to-melody, and melody-to-lyric, etc, a unified model for song composition has not yet been achieved. In this paper, we introduce SongComposer, a pioneering step towards a unified song composition model that can readily create symbolic lyrics and melodies following instructions. SongComposer is a music-specialized large language model (LLM) that, for the first time, integrates the capability of simultaneously composing lyrics and melodies into LLMs by leveraging three key innovations: 1) a flexible tuple format for word-level alignment of lyrics and melodies, 2) an extended tokenizer vocabulary for song notes, with scalar initialization based on musical knowledge to capture rhythm, and 3) a multi-stage pipeline that captures musical structure, starting with motif-level melody patterns and progressing to phrase-level structure for improved coherence. Extensive experiments demonstrate that SongComposer outperforms advanced LLMs, including GPT-4, in tasks such as lyric-to-melody generation, melody-to-lyric generation, song continuation, and text-to-song creation. Moreover, we will release SongCompose, a large-scale dataset for training, containing paired lyrics and melodies in Chinese and English.

Recent advancements in large language models (LLMs) have significantly enhanced their coding capabilities. However, existing benchmarks predominantly focused on simplified or isolated aspects of coding, such as single-file code generation or repository issue debugging, falling short of measuring the full spectrum of challenges raised by real-world programming activities. In this case study, we explore the performance of LLMs across the entire software development lifecycle with DevEval, encompassing stages including software design, environment setup, implementation, acceptance testing, and unit testing. DevEval features four programming languages, multiple domains, high-quality data collection, and carefully designed and verified metrics for each task. Empirical studies show that current LLMs, including GPT-4, fail to solve the challenges presented within DevEval. Our findings offer actionable insights for the future development of LLMs toward real-world programming applications.

Large Language Models (LLMs) have shown outstanding breakthroughs in code generation. Recent work improves code LLMs by training on synthetic data generated by some powerful LLMs, which can be challenging to scale due to the dependence on a teacher model and high generation costs. In this paper, we focus on synthesizing code data at scale and propose a Case2Code task by exploiting the expressiveness and correctness of programs. Case2Code is an inductive inference task that aims to infer underlying code implementations by observing input-output examples or program behaviors, By incorporating LLMs to generate program inputs, and executing the program with these inputs to obtain the program outputs, we can synthesize diverse and high-quality Case2Code data at scale for training and evaluating code LLMs. Experimental results show that case-to-code induction is challenging for current representative LLMs if they are untrained. Models trained with Case2Code improve performance not only on distribution case-to-code induction but also various coding-generation tasks, demonstrating the great potential of large-scale synthetic data and inductive learning.

Active real-time interaction with video LLMs introduces a new paradigm for human-computer interaction, where the model not only understands user intent but also responds while continuously processing streaming video on the fly. Unlike offline video LLMs, which analyze the entire video before answering questions, active real-time interaction requires three capabilities: 1) Perception: real-time video monitoring and interaction capturing. 2) Decision: raising proactive interaction in proper situations, 3) Reaction: continuous interaction with users. However, inherent conflicts exist among the desired capabilities. The Decision and Reaction require a contrary Perception scale and grain, and the autoregressive decoding blocks the real-time Perception and Decision during the Reaction. To unify the conflicted capabilities within a harmonious system, we present Dispider, a solution built on a Disentangled Perception, Decision, and Reaction framework. Dispider features a lightweight Proactive Streaming Video Processing module that tracks the video stream and identifies optimal moments for interaction. Once the interaction is triggered, an asynchronous Precise Interaction module provides detailed responses, while the processing module continues to monitor the video in the meantime. Our disentangled and asynchronous design ensures timely, contextually accurate, and computationally efficient responses, making Dispider ideal for active real-time interaction for long-duration video streams. Experiments prove that Dispider outperforms existing methods not only in its superior understanding of video content in conventional video QA settings, but also in proactive response capability and temporal awareness under the streaming setting. The code and model are released at https://github.com/Mark12Ding/Dispider.

Video inpainting, which aims to fill missing regions with visually coherent content, has emerged as a crucial technique for creative applications such as editing. While existing approaches achieve visual consistency or text-guided generation, they often struggle to balance coherence and creative diversity. In this work, we introduce VideoRepainter, a two-stage framework that allows users to inpaint a keyframe using established image-level techniques, then propagate the changes to other frames. Our approach can leverage state-of-the-art image models for keyframe manipulation, thereby easing the burden of the video-inpainting process. To this end, we integrate an image-to-video model with a symmetric condition mechanism to address ambiguity caused by direct mask downsampling. We further explore efficient strategies for mask synthesis and parameter tuning to reduce costs in data processing and model training. Evaluations demonstrate our method achieves superior results in both visual fidelity and content diversity compared to existing approaches, providing a practical solution for creative video manipulation. See our Project Page for more details.

The increasing demand for high-quality 3D assets across various industries necessitates efficient and automated 3D content creation. Despite recent advancements in 3D generative models, existing methods still face challenges with optimization speed, geometric fidelity, and the lack of assets for physically based rendering (PBR). In this paper, we introduce 3DTopia-XL, a scalable native 3D generative model designed to overcome these limitations. 3DTopia-XL leverages a novel primitive-based 3D representation, PrimX, which encodes detailed shape, albedo, and material field into a compact tensorial format, facilitating the modeling of high-resolution geometry with PBR assets. On top of the novel representation, we propose a generative framework based on Diffusion Transformer (DiT), which comprises 1) Primitive Patch Compression, 2) and Latent Primitive Diffusion. 3DTopia-XL learns to generate high-quality 3D assets from textual or visual inputs. Extensive qualitative and quantitative evaluations are conducted to demonstrate that 3DTopia-XL significantly outperforms existing methods in generating high-quality 3D assets with fine-grained textures and materials, efficiently bridging the quality gap between generative models and real-world applications.

The text-to-video (T2V) generation models, offering convenient visual creation, have recently garnered increasing attention. Despite their substantial potential, the generated videos may present artifacts, including structural implausibility, temporal inconsistency, and a lack of motion, often resulting in near-static video. In this work, we have identified a correlation between the disparity of temporal attention maps across different blocks and the occurrence of temporal inconsistencies. Additionally, we have observed that the energy contained within the temporal attention maps is directly related to the magnitude of motion amplitude in the generated videos. Based on these observations, we present ByTheWay, a training-free method to improve the quality of text-to-video generation without introducing additional parameters, augmenting memory or sampling time. Specifically, ByTheWay is composed of two principal components: 1) Temporal Self-Guidance improves the structural plausibility and temporal consistency of generated videos by reducing the disparity between the temporal attention maps across various decoder blocks. 2) Fourier-based Motion Enhancement enhances the magnitude and richness of motion by amplifying the energy of the map. Extensive experiments demonstrate that ByTheWay significantly improves the quality of text-to-video generation with negligible additional cost. Our code is available at: https://github.com/Bujiazi/ByTheWay.

Seamless integration of both aerial and street view images remains a significant challenge in neural scene reconstruction and rendering. Existing methods predominantly focus on single domain, limiting their applications in immersive environments, which demand extensive free view exploration with large view changes both horizontally and vertically. We introduce Horizon-Gs, a novel approach built upon Gaussian Splatting techniques, tackles the unified reconstruction and rendering for aerial and street views. Our method addresses the key challenges of combining these perspectives with a new training strategy, overcoming viewpoint discrepancies to generate high-fidelity scenes. We also curate a high-quality aerial-to-ground views dataset encompassing both synthetic and real-world scene to advance further research. Experiments across diverse urban scene datasets confirm the effectiveness of our method.

In large vision-language models (LVLMs), images serve as inputs that carry a wealth of information. As the idiom "A picture is worth a thousand words" implies, representing a single image in current LVLMs can require hundreds or even thousands of tokens, which thereby severely impacting the efficiency. Previous approaches have attempted to reduce the number of image tokens either before or within the early layers of LVLMs. However, these strategies inevitably result in the loss of crucial image information. To address this challenge, we conduct an empirical study revealing that all visual tokens are necessary for LVLMs in the shallow layers, and token redundancy progressively increases in the deeper layers. To this end, we propose PyramidDrop, a visual redundancy reduction strategy for LVLMs to boost their efficiency in both training and inference with neglectable performance loss. Specifically, we partition the LVLM into several stages and drop part of the image tokens at the end of each stage with a pre-defined ratio. The dropping is based on a lightweight similarity calculation with a negligible time overhead. Extensive experiments demonstrate that PyramidDrop can achieve over 40% training time reduction and 55% inference FLOPs acceleration on leading LVLMs like LLaVA-NeXT, maintaining comparable multi-modal performance. Besides, PyramidDrop can also serve as a plug-and-play strategy to accelerate inference in a free way, with better performance and lower inference cost than counterparts. Our code is available at https://github.com/Cooperx521/PyramidDrop.

To guarantee service quality in transformer based large language model (LLM) serving, it is essential to meet the latency constraints of both the prefill phase (measured by Time-to-First-Token, TTFT) and the decode phase (measured by Time-per-Output-Token, TPOT). Non-disaggregated serving places prefill and decode on the same worker, while disaggregated serving places the prefill and decode on isolated workers. However, no single architecture excels in both TTFT and TPOT metrics. After conducting a root cause analysis, we concluded that in disaggregated LLM serving, prefill execution has minimal interference with decode execution but result in high queuing times. In contrast, non-disaggregated LLM serving effectively reduces queuing times but introduces significant interference between prefills and decodes. In order to leverage the best aspects of both non-disaggregated and disaggregated LLM serving, we have designed and implemented Tropical. Tropical introduces an sevice-level objectives (SLO)-aware multiplexing strategy that balances the queuing time and the interference, enabling the LLM serving to achieve high TTFT and TPOT SLOs simultaneously. Our evaluation of real-world datasets reveals that Tropical outperforms both state-of-the-art non-disaggregated and disaggregated LLM serving systems, achieving up to $2.09 \times$ more requests within a 90% SLO attainment. Specially, compared to the disaggregated LLM serving system, Tropical improves P90 TTFT performance by $9 \times$ with only an 15% reduction in P90 TPOT. Against the non-disaggregated LLM serving systems, Tropical delivers a $2.8 \times$ performance improvement in P90 TPOT while maintaining the same P90 TTFT.

Fine-tuning multi-turn dialogue systems requires high-quality supervision but often suffers from degraded performance when exposed to low-quality data.Supervision errors in early turns can propagate across subsequent turns, undermining coherence and response quality.Existing methods typically address data quality via static prefiltering, which decouples quality control from training and fails to mitigate turn-level error propagation.In this context, we propose ReSURE (Regularizing Supervision UnREliability), an adaptive learning method that dynamically downweights unreliable supervision without explicit filtering.ReSURE estimates per-turn loss distributions using Welford's online statistics and reweights sample losses on the fly accordingly.Experiments on both singlesource and mixed-quality datasets show improved stability and response quality.Notably, ReSURE enjoys positive Spearman correlations (0.21 ∼ 1.0 across multiple benchmarks) between response scores and number of samples regardless of data quality, which potentially paves the way for utilizing large-scale data effectively.

The Segment Anything Model 2 (SAM 2) has emerged as a powerful foundation model for object segmentation in both images and videos, paving the way for various downstream video applications. The crucial design of SAM 2 for video segmentation is its memory module, which prompts object-aware memories from previous frames for current frame prediction. However, its greedy-selection memory design suffers from the "error accumulation" problem, where an errored or missed mask will cascade and influence the segmentation of the subsequent frames, which limits the performance of SAM 2 toward complex long-term videos. To this end, we introduce SAM2Long, an improved training-free video object segmentation strategy, which considers the segmentation uncertainty within each frame and chooses the video-level optimal results from multiple segmentation pathways in a constrained tree search manner. In practice, we maintain a fixed number of segmentation pathways throughout the video. For each frame, multiple masks are proposed based on the existing pathways, creating various candidate branches. We then select the same fixed number of branches with higher cumulative scores as the new pathways for the next frame. After processing the final frame, the pathway with the highest cumulative score is chosen as the final segmentation result. Benefiting from its heuristic search design, SAM2Long is robust toward occlusions and object reappearances, and can effectively segment and track objects for complex long-term videos. Notably, SAM2Long achieves an average improvement of 3.0 points across all 24 head-to-head comparisons, with gains of up to 5.3 points in J&F on long-term video object segmentation benchmarks such as SA-V and LVOS. The code is released at https://github.com/Mark12Ding/SAM2Long.

The Instruction Following (IF) ability measures how well Multi-modal Large Language Models (MLLMs) understand exactly what users are telling them and whether they are doing it right. Existing multimodal instruction following training data is scarce, the benchmarks are simple with atomic instructions, and the evaluation strategies are imprecise for tasks demanding exact output constraints. To address this, we present MM-IFEngine, an effective pipeline to generate high-quality image-instruction pairs. Our MM-IFEngine pipeline yields large-scale, diverse, and high-quality training data MM-IFInstruct-23k, which is suitable for Supervised Fine-Tuning (SFT) and extended as MM-IFDPO-23k for Direct Preference Optimization (DPO). We further introduce MM-IFEval, a challenging and diverse multi-modal instruction-following benchmark that includes (1) both compose-level constraints for output responses and perception-level constraints tied to the input images, and (2) a comprehensive evaluation pipeline incorporating both rule-based assessment and judge model. We conduct SFT and DPO experiments and demonstrate that fine-tuning MLLMs on MM-IFInstruct-23k and MM-IFDPO-23k achieves notable gains on various IF benchmarks, such as MM-IFEval (+10.2$\%$), MIA (+7.6$\%$), and IFEval (+12.3$\%$). We have fully open-sourced the datasets (both SFT and DPO), evaluation code and training scripts at https://github.com/SYuan03/MM-IFEngine.

Recent advances in video generation can produce realistic, minute-long single-shot videos with scalable diffusion transformers. However, real-world narrative videos require multi-shot scenes with visual and dynamic consistency across shots. In this work, we introduce Long Context Tuning (LCT), a training paradigm that expands the context window of pre-trained single-shot video diffusion models to learn scene-level consistency directly from data. Our method expands full attention mechanisms from individual shots to encompass all shots within a scene, incorporating interleaved 3D position embedding and an asynchronous noise strategy, enabling both joint and auto-regressive shot generation without additional parameters. Models with bidirectional attention after LCT can further be fine-tuned with context-causal attention, facilitating auto-regressive generation with efficient KV-cache. Experiments demonstrate single-shot models after LCT can produce coherent multi-shot scenes and exhibit emerging capabilities, including compositional generation and interactive shot extension, paving the way for more practical visual content creation. See https://guoyww.github.io/projects/long-context-video/ for more details.

The rapid advancements in generative technology have emerged as a double-edged sword. While offering powerful tools that enhance convenience, they also pose significant social concerns. As defenders, current synthetic image detection methods often lack artifact-level textual interpretability and are overly focused on image manipulation detection, and current datasets usually suffer from outdated generators and a lack of fine-grained annotations. In this paper, we introduce SynthScars, a high-quality and diverse dataset consisting of 12,236 fully synthetic images with human-expert annotations. It features 4 distinct image content types, 3 categories of artifacts, and fine-grained annotations covering pixel-level segmentation, detailed textual explanations, and artifact category labels. Furthermore, we propose LEGION (LEarning to Ground and explain for Synthetic Image detectiON), a multimodal large language model (MLLM)-based image forgery analysis framework that integrates artifact detection, segmentation, and explanation. Building upon this capability, we further explore LEGION as a controller, integrating it into image refinement pipelines to guide the generation of higher-quality and more realistic images. Extensive experiments show that LEGION outperforms existing methods across multiple benchmarks, particularly surpassing the second-best traditional expert on SynthScars by 3.31% in mIoU and 7.75% in F1 score. Moreover, the refined images generated under its guidance exhibit stronger alignment with human preferences. The code, model, and dataset will be released.

Reinforcement Fine-Tuning (RFT) in Large Reasoning Models like OpenAI o1 learns from feedback on its answers, which is especially useful in applications when fine-tuning data is scarce. Recent open-source work like DeepSeek-R1 demonstrates that reinforcement learning with verifiable reward is one key direction in reproducing o1. While the R1-style model has demonstrated success in language models, its application in multi-modal domains remains under-explored. This work introduces Visual Reinforcement Fine-Tuning (Visual-RFT), which further extends the application areas of RFT on visual tasks. Specifically, Visual-RFT first uses Large Vision-Language Models (LVLMs) to generate multiple responses containing reasoning tokens and final answers for each input, and then uses our proposed visual perception verifiable reward functions to update the model via the policy optimization algorithm such as Group Relative Policy Optimization (GRPO). We design different verifiable reward functions for different perception tasks, such as the Intersection over Union (IoU) reward for object detection. Experimental results on fine-grained image classification, few-shot object detection, reasoning grounding, as well as open-vocabulary object detection benchmarks show the competitive performance and advanced generalization ability of Visual-RFT compared with Supervised Fine-tuning (SFT). For example, Visual-RFT improves accuracy by $24.3\%$ over the baseline in one-shot fine-grained image classification with around 100 samples. In few-shot object detection, Visual-RFT also exceeds the baseline by $21.9$ on COCO's two-shot setting and $15.4$ on LVIS. Our Visual-RFT represents a paradigm shift in fine-tuning LVLMs, offering a data-efficient, reward-driven approach that enhances reasoning and adaptability for domain-specific tasks.

Generating human videos from a single image while ensuring high visual quality and precise control is a challenging task, especially in complex scenarios involving multiple individuals and interactions with objects. Existing methods, while effective for single-human cases, often fail to handle the intricacies of multi-identity interactions because they struggle to associate the correct pairs of human appearance and pose condition and model the distribution of 3D-aware dynamics. To address these limitations, we present \emph{Structural Video Diffusion}, a novel framework designed for generating realistic multi-human videos. Our approach introduces two core innovations: identity-specific embeddings to maintain consistent appearances across individuals and a structural learning mechanism that incorporates depth and surface-normal cues to model human-object interactions. Additionally, we expand existing human video dataset with 25K new videos featuring diverse multi-human and object interaction scenarios, providing a robust foundation for training. Experimental results demonstrate that Structural Video Diffusion achieves superior performance in generating lifelike, coherent videos for multiple subjects with dynamic and rich interactions, advancing the state of human-centric video generation. Code is available at https://github.com/zhenzhiwang/Multi-HumanVid

This paper aims to manipulate multi-entity 3D motions in video generation. Previous methods on controllable video generation primarily leverage 2D control signals to manipulate object motions and have achieved remarkable synthesis results. However, 2D control signals are inherently limited in expressing the 3D nature of object motions. To overcome this problem, we introduce 3DTrajMaster, a robust controller that regulates multi-entity dynamics in 3D space, given user-desired 6DoF pose (location and rotation) sequences of entities. At the core of our approach is a plug-and-play 3D-motion grounded object injector that fuses multiple input entities with their respective 3D trajectories through a gated self-attention mechanism. In addition, we exploit an injector architecture to preserve the video diffusion prior, which is crucial for generalization ability. To mitigate video quality degradation, we introduce a domain adaptor during training and employ an annealed sampling strategy during inference. To address the lack of suitable training data, we construct a 360-Motion Dataset, which first correlates collected 3D human and animal assets with GPT-generated trajectory and then captures their motion with 12 evenly-surround cameras on diverse 3D UE platforms. Extensive experiments show that 3DTrajMaster sets a new state-of-the-art in both accuracy and generalization for controlling multi-entity 3D motions. Project page: http://fuxiao0719.github.io/projects/3dtrajmaster

Large language models (LLMs) exhibit hallucinations (i.e., unfaithful or nonsensical information) when serving as AI assistants in various domains. Since hallucinations always come with truthful content in the LLM responses, previous factuality alignment methods that conduct response-level preference learning inevitably introduced noises during training. Therefore, this paper proposes a fine-grained factuality alignment method based on Direct Preference Optimization (DPO), called Mask-DPO. Incorporating sentence-level factuality as mask signals, Mask-DPO only learns from factually correct sentences in the preferred samples and prevents the penalty on factual contents in the not preferred samples, which resolves the ambiguity in the preference learning. Extensive experimental results demonstrate that Mask-DPO can significantly improve the factuality of LLMs responses to questions from both in-domain and out-of-domain datasets, although these questions and their corresponding topics are unseen during training. Only trained on the ANAH train set, the score of Llama3.1-8B-Instruct on the ANAH test set is improved from 49.19% to 77.53%, even surpassing the score of Llama3.1-70B-Instruct (53.44%), while its FactScore on the out-of-domain Biography dataset is also improved from 30.29% to 39.39%. We further study the generalization property of Mask-DPO using different training sample scaling strategies and find that scaling the number of topics in the dataset is more effective than the number of questions. We provide a hypothesis of what factual alignment is doing with LLMs, on the implication of this phenomenon, and conduct proof-of-concept experiments to verify it. We hope the method and the findings pave the way for future research on scaling factuality alignment.

Capturing geometric and material information from images remains a fundamental challenge in computer vision and graphics. Traditional optimization-based methods often require hours of computational time to reconstruct geometry, material properties, and environmental lighting from dense multi-view inputs, while still struggling with inherent ambiguities between lighting and material. On the other hand, learning-based approaches leverage rich material priors from existing 3D object datasets but face challenges with maintaining multi-view consistency. In this paper, we introduce IDArb, a diffusion-based model designed to perform intrinsic decomposition on an arbitrary number of images under varying illuminations. Our method achieves highly accurate and multi-view consistent estimation on surface normals and material properties. This is made possible through a novel cross-view, cross-domain attention module and an illumination-augmented, view-adaptive training strategy. Additionally, we introduce ARB-Objaverse, a new dataset that provides large-scale multi-view intrinsic data and renderings under diverse lighting conditions, supporting robust training. Extensive experiments demonstrate that IDArb outperforms state-of-the-art methods both qualitatively and quantitatively. Moreover, our approach facilitates a range of downstream tasks, including single-image relighting, photometric stereo, and 3D reconstruction, highlighting its broad applicability in realistic 3D content creation. Project website: https://lizb6626.github.io/IDArb/.

Visual preference alignment involves training Large Vision-Language Models (LVLMs) to predict human preferences between visual inputs. This is typically achieved by using labeled datasets of chosen/rejected pairs and employing optimization algorithms like direct preference optimization (DPO). Existing visual alignment methods, primarily designed for single-image scenarios, struggle to effectively handle the complexity of multi-image tasks due to the scarcity of diverse training data and the high cost of annotating chosen/rejected pairs. We present Multi-Image Augmented Direct Preference Optimization (MIA-DPO), a visual preference alignment approach that effectively handles multi-image inputs. MIA-DPO mitigates the scarcity of diverse multi-image training data by extending single-image data with unrelated images arranged in grid collages or pic-in-pic formats, significantly reducing the costs associated with multi-image data annotations. Our observation reveals that attention values of LVLMs vary considerably across different images. We use attention values to identify and filter out rejected responses the model may have mistakenly focused on. Our attention-aware selection for constructing the chosen/rejected pairs without relying on (i) human annotation, (ii) extra data, and (iii) external models or APIs. MIA-DPO is compatible with various architectures and outperforms existing methods on five multi-image benchmarks, achieving an average performance boost of 3.0% on LLaVA-v1.5 and 4.3% on the recent InternLM-XC2.5. Moreover, MIA-DPO has a minimal effect on the model's ability to understand single images.

Current efforts to learn scalable policies in robotic manipulation primarily fall into two categories: one focuses on "action," which involves behavior cloning from extensive collections of robotic data, while the other emphasizes "vision," enhancing model generalization by pre-training representations or generative models, also referred to as world models, using large-scale visual datasets. This paper presents an end-to-end paradigm that predicts actions using inverse dynamics models conditioned on the robot's forecasted visual states, named Predictive Inverse Dynamics Models (PIDM). By closing the loop between vision and action, the end-to-end PIDM can be a better scalable action learner. In practice, we use Transformers to process both visual states and actions, naming the model Seer. It is initially pre-trained on large-scale robotic datasets, such as DROID, and can be adapted to real-world scenarios with a little fine-tuning data. Thanks to large-scale, end-to-end training and the continuous synergy between vision and action at each execution step, Seer significantly outperforms state-of-the-art methods across both simulation and real-world experiments. It achieves improvements of 13% on the LIBERO-LONG benchmark, 22% on CALVIN ABC-D, and 43% in real-world tasks. Notably, it demonstrates superior generalization for novel objects, lighting conditions, and environments under high-intensity disturbances. Code and models will be publicly available.

With the rapid development of AI-generated content, the future internet may be inundated with synthetic data, making the discrimination of authentic and credible multimodal data increasingly challenging. Synthetic data detection has thus garnered widespread attention, and the performance of large multimodal models (LMMs) in this task has attracted significant interest. LMMs can provide natural language explanations for their authenticity judgments, enhancing the explainability of synthetic content detection. Simultaneously, the task of distinguishing between real and synthetic data effectively tests the perception, knowledge, and reasoning capabilities of LMMs. In response, we introduce LOKI, a novel benchmark designed to evaluate the ability of LMMs to detect synthetic data across multiple modalities. LOKI encompasses video, image, 3D, text, and audio modalities, comprising 18K carefully curated questions across 26 subcategories with clear difficulty levels. The benchmark includes coarse-grained judgment and multiple-choice questions, as well as fine-grained anomaly selection and explanation tasks, allowing for a comprehensive analysis of LMMs. We evaluated 22 open-source LMMs and 6 closed-source models on LOKI, highlighting their potential as synthetic data detectors and also revealing some limitations in the development of LMM capabilities. More information about LOKI can be found at https://opendatalab.github.io/LOKI/.

Mixture-of-Experts (MoE) models face deployment challenges due to their large parameter counts and computational demands. We explore quantization for MoE models and highlight two key insights: 1) linear blocks exhibit varying quantization sensitivity, and 2) divergent expert activation frequencies create heterogeneous computational characteristics. Based on these observations, we introduce MxMoE, a mixed-precision optimization framework for MoE models that considers both algorithmic and system perspectives. MxMoE navigates the design space defined by parameter sensitivity, expert activation dynamics, and hardware resources to derive efficient mixed-precision configurations. Additionally, MxMoE automatically generates optimized mixed-precision GroupGEMM kernels, enabling parallel execution of GEMMs with different precisions. Evaluations show that MxMoE outperforms existing methods, achieving 2.4 lower Wikitext-2 perplexity than GPTQ at 2.25-bit and delivering up to 3.4x speedup over full precision, as well as up to 29.4% speedup over uniform quantization at equivalent accuracy with 5-bit weight-activation quantization. Our code is available at https://github.com/cat538/MxMoE.

Text-to-song generation, the task of creating vocals and accompaniment from textual inputs, poses significant challenges due to domain complexity and data scarcity. Existing approaches often employ multi-stage generation procedures, leading to cumbersome training and inference pipelines, as well as suboptimal overall generation quality due to error accumulation across stages. In this paper, we propose SongGen, a fully open-source, single-stage auto-regressive transformer designed for controllable song generation. The proposed model facilitates fine-grained control over diverse musical attributes, including lyrics and textual descriptions of instrumentation, genre, mood, and timbre, while also offering an optional three-second reference clip for voice cloning. Within a unified auto-regressive framework, SongGen supports two output modes: mixed mode, which generates a mixture of vocals and accompaniment directly, and dual-track mode, which synthesizes them separately for greater flexibility in downstream applications. We explore diverse token pattern strategies for each mode, leading to notable improvements and valuable insights. Furthermore, we design an automated data preprocessing pipeline with effective quality control. To foster community engagement and future research, we will release our model weights, training code, annotated data, and preprocessing pipeline. The code is available at https://github.com/LiuZH-19/SongGen.

While Rotary Position Embedding (RoPE) and its variants are widely adopted for their long-context capabilities, the extension of the 1D RoPE to video, with its complex spatio-temporal structure, remains an open challenge. This work first introduces a comprehensive analysis that identifies four key characteristics essential for the effective adaptation of RoPE to video, which have not been fully considered in prior work. As part of our analysis, we introduce a challenging V-NIAH-D (Visual Needle-In-A-Haystack with Distractors) task, which adds periodic distractors into V-NIAH. The V-NIAH-D task demonstrates that previous RoPE variants, lacking appropriate temporal dimension allocation, are easily misled by distractors. Based on our analysis, we introduce VideoRoPE, with a 3D structure designed to preserve spatio-temporal relationships. VideoRoPE features low-frequency temporal allocation to mitigate periodic oscillations, a diagonal layout to maintain spatial symmetry, and adjustable temporal spacing to decouple temporal and spatial indexing. VideoRoPE consistently surpasses previous RoPE variants, across diverse downstream tasks such as long video retrieval, video understanding, and video hallucination. Our code and model weights will be publicly released.

Vision-language instruction-tuning models have recently achieved significant performance improvements. In this work, we discover that large-scale 3D parallel training on those models leads to an imbalanced computation load across different devices. The vision and language parts are inherently heterogeneous: their data distribution and model architecture differ significantly, which affects distributed training efficiency. To address this issue, we rebalance the computational load from data, model, and memory perspectives, achieving more balanced computation across devices. Specifically, for the data, instances are grouped into new balanced mini-batches within and across devices. A search-based method is employed for the model to achieve a more balanced partitioning. For memory optimization, we adaptively adjust the re-computation strategy for each partition to utilize the available memory fully. These three perspectives are not independent but are closely connected, forming an omniverse balanced training framework. Extensive experiments are conducted to validate the effectiveness of our method. Compared with the open-source training code of InternVL-Chat, training time is reduced greatly, achieving about 1.8$\times$ speed-up. Our method’s efficacy and generalizability are further validated across various models and datasets. Codes will be released at https://github.com/ModelTC/OmniBal.

Text-to-image (T2I) diffusion/flow models have drawn considerable attention recently due to their remarkable ability to deliver flexible visual creations. Still, high-resolution image synthesis presents formidable challenges due to the scarcity and complexity of high-resolution content. Recent approaches have investigated training-free strategies to enable high-resolution image synthesis with pre-trained models. However, these techniques often struggle with generating high-quality visuals and tend to exhibit artifacts or low-fidelity details, as they typically rely solely on the endpoint of the low-resolution sampling trajectory while neglecting intermediate states that are critical for preserving structure and synthesizing finer detail. To this end, we present HiFlow, a training-free and model-agnostic framework to unlock the resolution potential of pre-trained flow models. Specifically, HiFlow establishes a virtual reference flow within the high-resolution space that effectively captures the characteristics of low-resolution flow information, offering guidance for high-resolution generation through three key aspects: initialization alignment for low-frequency consistency, direction alignment for structure preservation, and acceleration alignment for detail fidelity. By leveraging such flow-aligned guidance, HiFlow substantially elevates the quality of high-resolution image synthesis of T2I models and demonstrates versatility across their personalized variants. Extensive experiments validate HiFlow's capability in achieving superior high-resolution image quality over state-of-the-art methods.

Large language models (LLMs) have demonstrated exceptional performance in reasoning tasks such as mathematics and coding, matching or surpassing human capabilities. However, these impressive reasoning abilities face significant challenges in specialized domains. Taking Go as an example, although AlphaGo has established the high performance ceiling of AI systems in Go, mainstream LLMs still struggle to reach even beginner-level proficiency, let alone perform natural language reasoning. This performance gap between general-purpose LLMs and domain experts is significantly limiting the application of LLMs on a wider range of domain-specific tasks. In this work, we aim to bridge the divide between LLMs' general reasoning capabilities and expert knowledge in domain-specific tasks. We perform mixed fine-tuning with structured Go expertise and general long Chain-of-Thought (CoT) reasoning data as a cold start, followed by reinforcement learning to integrate expert knowledge in Go with general reasoning capabilities. Through this methodology, we present LoGos, a powerful LLM that not only maintains outstanding general reasoning abilities, but also conducts Go gameplay in natural language, demonstrating effective strategic reasoning and accurate next-move prediction. LoGos achieves performance comparable to human professional players, substantially surpassing all existing LLMs. Through this work, we aim to contribute insights on applying general LLM reasoning capabilities to specialized domains. We will release the first large-scale Go dataset for LLM training, the first LLM Go evaluation benchmark, and the first general LLM that reaches human expert-level performance in Go.

Enhancing large vision-language models (LVLMs) with visual slow-thinking reasoning is crucial for solving complex multimodal tasks. However, since LVLMs are mainly trained with vision-language alignment, it is difficult to adopt on-policy reinforcement learning (RL) to develop the slow thinking ability because the rollout space is restricted by its initial abilities. Off-policy RL offers a way to go beyond the current policy, but directly distilling trajectories from external models may cause visual hallucinations due to mismatched visual perception abilities across models. To address these issues, this paper proposes **SOPHIA**, a simple and scalable **S**emi-**O**ff-**P**olicy RL for vision-language slow-t**HI**nking re**A**soning. SOPHIA builds a semi-off-policy behavior model by combining on-policy visual understanding from a trainable LVLM with off-policy slow-thinking reasoning from a language model, assigns outcome-based rewards to reasoning, and propagates visual rewards backward. Then LVLM learns slow-thinking reasoning ability from the obtained reasoning trajectories using propagated rewards via off-policy RL algorithms. Extensive experiments with InternVL2.5 and InternVL3.0 with 8B and 38B sizes show the effectiveness of SOPHIA. Notably, SOPHIA improves InternVL3.0-38B by 8.50\% in average, reaching state-of-the-art performance among open-source LVLMs on multiple multimodal reasoning benchmarks, and even outperforms some closed-source models (e.g., GPT-4.1) on the challenging MathVision and OlympiadBench, achieving 49.08\% and 49.95\% pass@1 accuracy, respectively. Analysis shows SOPHIA outperforms supervised fine-tuning and direct on-policy RL methods, offering a better policy initialization for further on-policy training.

$360^\circ$ videos offer a hyper-immersive experience that allows the viewers to explore a dynamic scene from full 360 degrees. To achieve more accessible and personalized content creation in $360^\circ$ video format, we seek to lift standard perspective videos into $360^\circ$ equirectangular videos. To this end, we introduce **Imagine360**, the first perspective-to-$360^\circ$ video generation framework that creates high-quality $360^\circ$ videos with rich and diverse motion patterns from video anchors. Imagine360 learns fine-grained spherical visual and motion patterns from limited $360^\circ$ video data with several key designs. **1)** Firstly we adopt the dual-branch design, including a perspective and a panorama video denoising branch to provide local and global constraints for $360^\circ$ video generation, with motion module and spatial LoRA layers fine-tuned on $360^\circ$ videos. **2)** Additionally, an antipodal mask is devised to capture long-range motion dependencies, enhancing the reversed camera motion between antipodal pixels across hemispheres. **3)** To handle diverse perspective video inputs, we propose rotation-aware designs that adapt to varying video masking due to changing camera poses across frames. **4)** Lastly, we introduce a new 360 video dataset featuring 10K high-quality, trimmed 360 video clips with structured motion to facilitate training. Extensive experiments show Imagine360 achieves superior graphics quality and motion coherence with our curated dataset among state-of-the-art $360^\circ$ video generation methods. We believe Imagine360 holds promise for advancing personalized, immersive $360^\circ$ video creation.

Emerging world models autoregressively generate video frames in response to actions, such as camera movements and text prompts, among other control signals. Due to limited temporal context window sizes, these models often struggle to maintain scene consistency during revisits, leading to severe forgetting of previously generated environments. Inspired by the mechanisms of human memory, we introduce a novel framework to enhancing long-term consistency of video world models through a geometry-grounded long-term spatial memory. Our framework includes mechanisms to store and retrieve information from the long-term spatial memory and we curate custom datasets to train and evaluate world models with explicitly stored 3D memory mechanisms. Our evaluations show improved quality, consistency, and context length compared to relevant baselines, paving the way towards long-term consistent world generation.

Training Long-Context Large Language Models (LLMs) is challenging, as hybrid training with long-context and short-context data often leads to workload imbalances. Existing works mainly use data packing to alleviate this issue, but fail to consider imbalanced attention computation and wasted communication overhead. This paper proposes Hierarchical Balance Packing (HBP), which designs a novel batch-construction method and training recipe to address those inefficiencies. In particular, the HBP constructs multi-level data packing groups, each optimized with a distinct packing length. It assigns training samples to their optimal groups and configures each group with the most effective settings, including sequential parallelism degree and gradient checkpointing configuration. To effectively utilize multi-level groups of data, we design a dynamic training pipeline specifically tailored to HBP, including curriculum learning, adaptive sequential parallelism, and stable loss. Our extensive experiments demonstrate that our method significantly reduces training time over multiple datasets and open-source models while maintaining strong performance. For the largest DeepSeek-V2 (236B) MoE model, our method speeds up the training by 2.4$\times$ with competitive performance. Codes will be released at https://github.com/ModelTC/HBP.

Despite rapid advances in 3D content generation, quality assessment for the generated 3D assets remains challenging.Existing methods mainly rely on image-based metrics and operate solely at the object level, limiting their ability to capture spatial Despite rapid advances in 3D content generation, quality assessment for the generated 3D assets remains challenging.Existing methods mainly rely on image-based metrics and operate solely at the object level, limiting their ability to capture spatial coherence, material authenticity, and high-fidelity local details.1) To address these challenges, we introduce Hi3DEval, a hierarchical evaluation framework tailored for 3D generative content. It combines both object-level and part-level evaluation, enabling holistic assessments across multiple dimensions as well as fine-grained quality analysis. Additionally, we extend texture evaluation beyond aesthetic appearance by explicitly assessing material realism, focusing on attributes such as albedo, saturation, and metallicness. 2) To support this framework, we construct Hi3DBench, a large-scale dataset comprising diverse 3D assets and high-quality annotations, accompanied by a reliable multi-agent annotation pipeline.We further propose a 3D-aware automated scoring system based on hybrid 3D representations. Specifically, we leverage video-based representations for object-level and material-subject evaluations to enhance modeling of spatio-temporal consistency and employ pretrained 3D features for part-level perception.Extensive experiments demonstrate that our approach outperforms existing image-based metrics in modeling 3D characteristics and achieves superior alignment with human preference, providing a scalable alternative to manual evaluations.

We introduce AnySplat, a feed-forward network for novel-view synthesis from uncalibrated image collections. In contrast to traditional neural-rendering pipelines that demand known camera poses and per-scene optimization, or recent feed-forward methods that buckle under the computational weight of dense views—our model predicts everything in one shot. A single forward pass yields a set of 3D Gaussian primitives encoding both scene geometry and appearance, and the corresponding camera intrinsics and extrinsics for each input image. This unified design scales effortlessly to casually captured, multi-view datasets without any pose annotations. In extensive zero-shot evaluations, AnySplat matches the quality of pose-aware baselines in both sparse- and dense-view scenarios while surpassing existing pose-free approaches. Moreover, it greatly reduces rendering latency compared to optimization-based neural fields, bringing real-time novel-view synthesis within reach for unconstrained capture settings. Project page: https://city-super.github.io/anysplat/.

We present SS4D, a native 4D generative model that synthesizes dynamic 3D objects directly from monocular video. Unlike prior approaches that construct 4D representations by optimizing over 3D or video generative models, we train a generator directly on 4D data, achieving high fidelity, temporal coherence, and structural consistency. At the core of our method is a compressed set of structured spacetime latents. Specifically, (1) To address the scarcity of 4D training data, we build on a pre-trained single-image-to-3D model, preserving strong spatial consistency. (2) Temporal consistency is enforced by introducing dedicated temporal layers that reason across frames. (3) To support efficient training and inference over long video sequences, we compress the latent sequence along the temporal axis using factorized 4D convolutions and temporal downsampling blocks. In addition, we employ a carefully designed training strategy to enhance robustness against occlusion and motion blur, leading to high-quality generation. Extensive experiments show that SS4D produces spatio-temporally consistent 4D objects with superior quality and efficiency, significantly outperforming state-of-the-art methods on both synthetic and real-world datasets.

Zehui Chen, Weihua Du, Wenwei Zhang, Kuikun Liu, Jiangning Liu, Miao Zheng, Jingming Zhuo, Songyang Zhang, Dahua Lin, Kai Chen, Feng Zhao. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Reducing the 'hallucination' problem of Large Language Models (LLMs) is crucial for their wide applications.A comprehensive and finegrained measurement of the hallucination is the first key step for the governance of this issue but is under-explored in the community.Thus, we present ANAH, a bilingual dataset that offers ANalytical Annotation of Hallucinations in LLMs within Generative Question Answering.Each answer sentence in our dataset undergoes rigorous annotation, involving the retrieval of a reference fragment, the judgment of the hallucination type, and the correction of hallucinated content.ANAH consists of ∼12k sentence-level annotations for ∼4.3kLLM responses covering over 700 topics, constructed by a human-in-the-loop pipeline.Thanks to the fine granularity of the hallucination annotations, we can quantitatively confirm that the hallucinations of LLMs progressively accumulate in the answer and use ANAH to train and evaluate hallucination annotators.We conduct extensive experiments on studying generative and discriminative annotators and show that, although current open-source LLMs have difficulties in fine-grained hallucination annotation, the generative annotator trained with ANAH can surpass all open-source LLMs and GPT-3.5, obtain performance competitive with GPT-4, and exhibits better generalization ability on unseen questions. 1

Chenyu Shi, Xiao Wang, Qiming Ge, Songyang Gao, Xianjun Yang, Tao Gui, Qi Zhang, Xuanjing Huang, Xun Zhao, Dahua Lin. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Yu Sun, Keyu Chen, Shujie Wang, Peiji Li, Qipeng Guo, Hang Yan, Xipeng Qiu, Xuanjing Huang, Dahua Lin. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

As instruction-tuned large language models (LLMs) evolve, aligning pretrained foundation models presents increasing challenges.Existing alignment strategies, which typically leverage diverse and high-quality data sources, often overlook the intrinsic uncertainty of tasks, learning all data samples equally.This may lead to suboptimal data efficiency and model performance.In response, we propose uncertainty-aware learning (UAL) to improve the model alignment of different task scenarios, by introducing the sample uncertainty (elicited from more capable LLMs).We implement UAL by a simple fashion -adaptively setting the label smoothing value of training according to the uncertainty of individual samples.Analysis shows that our UAL indeed facilitates better token clustering in the feature space, validating our hypothesis.Extensive experiments on widely used benchmarks demonstrate that our UAL significantly and consistently outperforms standard supervised fine-tuning.Notably, LLMs aligned in a mixed scenario have achieved an average improvement of 10.62% on high-entropy tasks (i.e., AlpacaEval leaderboard), and 1.81% on complex low-entropy tasks (i.e., MetaMath and GSM8K).

Intelligent technology is driving the communication industry to a higher stage of autonomy. To enable the large-scale deployment of Autonomous Networks (ANs), We design a knowledge-driven self-healing dual-loop architecture throughout the fault lifecycle, Then we propose an optimization model that aims to minimize the loss of services operation. Simulations are conducted in a programmable network to validate the loop.

The high computational and memory requirements of generative large language models (LLMs) make it challenging to serve them cheaply. This paper aims to reduce the monetary cost for serving LLMs by leveraging preemptible GPU instances on modern clouds, which offer accesses to spare GPU resources at a much cheaper price than regular instances but may be preempted by the cloud provider at any time. Serving LLMs on preemptible instances requires addressing challenges induced by frequent instance preemptions and the necessity of migrating instances to handle the preemptions.

Contrastive Language-Image Pre-training (CLIP) plays an essential role in extracting valuable content information from images across diverse tasks. It aligns textual and visual modalities to comprehend the entire image, including all the details, even those irrelevant to specific tasks. How-ever, for a finer understanding and controlled editing of images, it becomes crucial to focus on specific regions of interest, which can be indicated as points, masks, or boxes by humans or perception models. To fulfill the require we introduce Alpha-CLIp, an enhanced version of CLIP with an auxiliary alpha channel to suggest attentive regions and fine-tuned with constructed millions of RGBA region-text pairs. Alpha-CLIP not only preserves the visual recognition ability of CLIP but also enables precise control over the emphasis of image contents. It demonstrates effectiveness in various tasks, including but not limited to open-world recognition, multimodal large language models, and conditional 2D / 3D generation. It has a strong potential to serve as a versatile tool for image-related tasks. Our project is with codes and models available is linked to https://aleafy.github.io/alpha-clip/.

Text-driven video generation witnesses rapid progress. However, merely using text prompts is not enough to depict the desired subject appearance that accurately aligns with users' intents, especially for customized content creation. In this paper, we study the task of video generation with image prompts, which provide more accurate and direct content control beyond the text prompts. Specifically, we propose a feed-forward framework VideoBooth, with two dedicated designs: 1) We propose to embed image prompts in a coarse-to-fine manner. Coarse visual embeddings from image encoder provide high-level encodings of image prompts, while fine visual embeddings from the proposed attention injection module provide multi-scale and detailed encoding of image prompts. These two complementary embeddings can faithfully capture the desired appearance. 2) In the attention injection module at fine level, multi-scale image prompts are fed into different cross-frame attention layers as additional keys and values. This extra spatial in-formation refines the details in the first frame and then it is propagated to the remaining frames, which maintains temporal consistency. Extensive experiments demonstrate that Video Booth achieves state-of-the-art performance in gener-ating customized high-quality videos with subjects specified in image prompts. Notably, VideoBooth is a generalizable framework where a single model works for a wide range of image prompts with only feed-forward passes.

Neural rendering methods have significantly advanced photo-realistic 3D scene rendering in various academic and industrial applications. The recent 3D Gaussian Splatting method has achieved the state-of-the-art rendering quality and speed combining the benefits of both primitive-based representations and volumetric representations. However, it often leads to heavily redundant Gaussians that try to fit every training view, neglecting the underlying scene ge-ometry. Consequently, the resulting model becomes less robust to significant view changes, texture-less area and lighting effects. We introduce Scaffold-GS, which uses an-chor points to distribute local 3D Gaussians, and predicts their attributes on-the-fly based on viewing direction and distance within the view frustum. Anchor growing and pruning strategies are developed based on the importance of neural Gaussians to reliably improve the scene cover-age. We show that our method effectively reduces redun-dant Gaussians while delivering high-quality rendering. We also demonstrates an enhanced capability to accommodate scenes with varying levels-of-detail and view-dependent ob-servations, without sacrificing the rendering speed. Project page: https://city-super.github.iolscaffold-gsl.

Multimodal large language models (MLLMs) have gained significant attention due to their strong multimodal understanding capability. However, existing works rely heavily on modality-specific encoders, which usually differ in architecture and are limited to common modalities. In this paper, we present OneLLM, an MLLM that aligns eight modalities to language using a unified framework. We achieve this through a unified multimodal encoder and a progressive multimodal alignment pipeline. In detail, we first train an image projection module to connect a vision encoder with LLM. Then, we build a universal projection module (UPM) by mixing multiple image projection modules and dynamic routing. Finally, we progressively align more modalities to LLM with the UPM. To fully leverage the potential of OneLLM in following instructions, we also curated a comprehensive multimodal instruction dataset, including 2M items from image, audio, video, point cloud, depth/normal map, IMU and fMRI brain activity. OneLLM is evaluated on 25 diverse benchmarks, encompassing tasks such as multimodal captioning, question answering and reasoning, where it delivers excellent performance. Code, data, model and online demo are available at https://github.com/csuhan/OneLLM.

Hallucination, posed as a pervasive challenge of multi-modal large language models (MLLMs), has significantly impeded their real-world usage that demands precise judgment. Existing methods mitigate this issue with either training with specific designed data or inferencing with external knowledge from other sources, incurring inevitable additional costs. In this paper, we present OPERA, a novel MLLM decoding method grounded in an Over-trust Penalty and a Retrospection-Allocation strategy, serving as a nearly free lunch to alleviate the hallucination issue without additional data, knowledge, or training. Our approach begins with an interesting observation that, most hallucinations are closely tied to the knowledge aggregation patterns manifested in the self-attention matrix, i.e., MLLMs tend to generate new tokens by focusing on a few summary tokens, but not all the previous tokens. Such partial overtrust inclination results in the neglecting of image tokens and describes the image content with hallucination. Based on the observation, OPERA introduces a penalty term on the model logits during the beam-search decoding to mitigate the over-trust issue, along with a rollback strategy that retrospects the presence of summary tokens in the previously generated tokens, and re-allocate the token selection if necessary. With extensive experiments, OPERA shows significant hallucination-mitigating performance on different MLLMs and metrics, proving its effectiveness and generality. Our code is at: https://github.com/shikiw/OPERA.

Video generation has witnessed significant advance-ments, yet evaluating these models remains a challenge. A comprehensive evaluation benchmark for video generation is indispensable for two reasons: 1) Existing metrics do not fully align with human perceptions; 2) An ideal eval-uation system should provide insights to inform future de-velopments of video generation. To this end, we present VBench, a comprehensive benchmark suite that dissects “video generation quality” into specific, hierarchical, and disentangled dimensions, each with tailored prompts and evaluation methods. VBench has three appealing proper-ties: 1) Comprehensive Dimensions: VBench comprises 16 dimensions in video generation (e.g., subject identity in-consistency, motion smoothness, temporal flickering, and spatial relationship, etc.). The evaluation metrics with fine-grained levels reveal individual models' strengths and weaknesses. 2) Human Alignment: We also provide a dataset of human preference annotations to validate our benchmarks' alignment with human perception, for each evaluation dimension respectively. 3) Valuable Insights: We look into current models' ability across various evaluation dimensions, and various content types. We also investi-gate the gaps between video and image generation models. We will open-source VBench, including all prompts, evaluation methods, generated videos, and human preference an-notations, and also include more video generation models in VBench to drive forward the field of video generation.

In the evolving landscape of digital media and video pro-duction, the precise manipulation and reproduction of visual elements like camera movements and character actions are highly desired. Existing SLAM methods face limitations in dynamic scenes and human pose estimation often focuses on 2D projections, neglecting 3D statuses. To address these is-sues, we first introduce a reverse filming behavior estimation technique. It optimizes camera trajectories by leveraging NeRF as a differentiable renderer and refining SMPL tracks. We then introduce a cinematic transfer pipeline that is able to transfer various shot types to a new 2D video or a 3D vir-tual environment. The incorporation of 3D engine workflow enables superior rendering and control abilities, which also achieves a higher rating in the user study.

Scene graph generation (SGG) aims to parse a visual scene into an intermediate graph representation for downstream reasoning tasks. Despite recent advancements, existing methods struggle to generate scene graphs with novel visual relation concepts. To address this challenge, we introduce a new open-vocabulary SGG framework based on sequence generation. Our framework leverages vision-language pre-trained models (VLM) by incorporating an image-to-graph generation paradigm. Specifically, we generate scene graph sequences via image-to-text generation with VLM and then construct scene graphs from these sequences. By doing so, we harness the strong capabilities of VLM for open-vocabulary SGG and seamlessly integrate explicit relational modeling for enhancing the VL tasks. Experimental results demonstrate that our design not only achieves superior performance with an open vocabulary but also enhances downstream vision-language task performance through explicit relation modeling knowledge.

Realistic 3D human generationfrom text prompts is a de-sirable yet challenging task. Existing methods optimize 3D representations like mesh or neural fields via score distil-lation sampling (SDS), which suffers from inadequate fine details or excessive training time. In this paper, we pro-pose an efficient yet effective framework, HumanGaussian, that generates high-quality 3D humans with fine-grained geometry and realistic appearance. Our key insight is that 3D Gaussian Splatting is an efficient renderer with peri-odic Gaussian shrinkage or growing, where such adaptive density control can be naturally guided by intrinsic human structures. Specifically, 1) we first propose a Structure-Aware SDS that simultaneously optimizes human appear-ance and geometry. The multi-modal score function from both RGB and depth space is leveraged to distill the Gaus-sian densification and pruning process. 2) Moreover, we devise an Annealed Negative Prompt Guidance by decom-posing SDS into a noisier generative score and a cleaner classifier score, which well addresses the over-saturation issue. The floating artifacts are further eliminated based on Gaussian size in a prune-only phase to enhance generation smoothness. Extensive experiments demonstrate the supe-rior efficiency and competitive quality of our framework, rendering vivid 3D humans under diverse scenarios.

Multimodal Large Language Models (MLLMs) have excelled in 2D image-text comprehension and image generation. Still, their understanding of the 3D world needs to be improved, limiting progress in 3D language understanding and generation. To solve this problem, we introduce GPT4Point, an innovative, groundbreaking point-language multimodal model explicitly designed for unified 3D object understanding and generation within the MLLMframework. GPT4Point, as a powerful 3D MLLM, can seamlessly execute point-text reference tasks such as point-cloud captioning and Q&A. Additionally, GPT4Point is equipped with advanced capabilities for controllable 3D generation, and it can get high-quality results through a low-quality point-text feature that maintains geometric shapes and colors. We develop Pyramid-XL, a point-language dataset annotation engine, to support the expansive needs of 3D object-text pairs. It constructs a large-scale database of over 1M objects of varied text granularity levels from the Objaverse-XL dataset, essential for training GPT4Point. A comprehensive benchmark has been proposed to evaluate 3D point-language understanding capabilities. In extensive evaluations, GPT4Point has demonstrated superior performance in understanding and generation.

In the realm of computer vision and robotics, embodied agents are expected to explore their environment and carry out human instructions. This necessitates the ability to fully understand 3D scenes given their first-person observations and contextualize them into language for interaction. However, traditional research focuses more on scene-level input and output setups from a global view. To address the gap, we introduce EmbodiedScan, a multi-modal, ego-centric 3D perception dataset and benchmark for holistic 3D scene understanding. It encompasses over 5k scans encapsulating 1M ego-centric RGB-D views, 1M language prompts, 160k 3D-oriented boxes spanning over 760 categories, some of which partially align with LVIS, and dense semantic occupancy with 80 common categories. Building upon this database, we introduce a baseline framework named Embodied Perceptron. It is capable of processing an arbitrary number of multi-modal inputs and demonstrates remarkable 3D perception capabilities, both within the two series of benchmarks we set up, i.e., fundamental 3D perception tasks and language-grounded tasks, and in the wild.

Despite recent advances in text-to-3D generative methods, there is a notable absence of reliable evaluation metrics. Existing metrics usually focus on a single criterion each, such as how well the asset aligned with the input text. These metrics lack the flexibility to generalize to different evaluation criteria and might not align well with human preferences. Conducting user preference studies is an alternative that offers both adaptability and human-aligned results. User studies, however, can be very ex-pensive to scale. This paper presents an automatic, ver-satile, and human-aligned evaluation metric for text-to-3D generative models. To this end, we first develop a prompt generator using GPT-4V to generate evaluating prompts, which serve as input to compare text-to-3D models. We further design a method instructing GPT-4V to compare two 3D assets according to user-defined crite-ria. Finally, we use these pairwise comparison results to assign these models Elo ratings. Experimental results suggest our metric strongly aligns with human preference across different evaluation criteria. Our code is available at https://github.com/3DTopia/GPTEval3D.

We are witnessing significant breakthroughs in the tech-nology for generating 3D objects from text. Existing approaches either leverage large text-to-image models to optimize a 3D representation or train 3D generators on object-centric datasets. Generating entire scenes, however, remains very challenging as a scene contains multiple 3D objects, diverse and scattered. In this work, we introduce SceneWiz3D - a novel approach to synthesize high-fidelity 3D scenes from text. We marry the locality of objects with globality of scenes by introducing a hybrid 3D representation - explicit for objects and implicit for scenes. Remarkably, an object, being represented explicitly, can be either generated from text using conventional text-to-3D approaches, or provided by users. To configure the layout of the scene and automatically place objects, we apply the Particle Swarm Optimization technique during the optimization process. Furthermore, it is difficult for certain parts of the scene (e.g., corners, occlusion) to receive multi-view supervision, leading to inferior geometry. We incor-porate an RGBD panorama diffusion model to mitigate it, resulting in high-quality geometry. Extensive evaluation supports that our approach achieves superior quality over previous approaches, enabling the generation of detailed and view-consistent 3D scenes. Our project website is at https://zqh0253.github.io/SceneWiz3D/.

As deep learning empowers various fields, many domain-specific non-neural network operators have been proposed to improve the accuracy of deep learning models. Researchers often use the imperative programming diagram (PyTorch) to express these new operators, leaving the fusion optimization of these operators to deep learning compilers. Unfortunately, the inherent side effects introduced by imperative tensor programs, especially tensor-level mutations, often make optimization extremely difficult. Previous works either fail to eliminate the side effects of tensor-level mutations or require programmers to manually analyze and transform them. In this paper, we present a holistic functionalization approach (TensorSSA) to optimizing imperative tensor programs beyond control flow boundaries. We first introduce TensorSSA intermediate representation for removing tensor-level mutation and expanding the scope and ability of operator fusion. Based on TensorSSA IR, we propose a TensorSSA conversion algorithm that performs functionalization crossing the boundary of control flow. TensorSSA achieves a 1.79X (1.34X on average) speedup in representative deep learning tasks than state-of-the-art works.

Zhiyuan Zeng, Qipeng Guo, Zhaoye Fei, Zhangyue Yin, Yunhua Zhou, Linyang Li, Tianxiang Sun, Hang Yan, Dahua Lin, Xipeng Qiu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Recent studies have demonstrated the significant potential of Large Language Models (LLMs) in generating Register Transfer Level (RTL) code, with notable advancements showcased by commercial models such as GPT-4 and Claude3-Opus. However, these proprietary LLMs often raise concerns regarding privacy and security. While open-source LLMs offer solutions to these concerns, they typically underperform commercial models in RTL code generation tasks, primarily due to the scarcity of high-quality open-source RTL datasets. To address this challenge, we introduce OriGen, a fully open-source framework that incorporates self-reflection capabilities and a novel dataset augmentation methodology for generating high-quality, large-scale RTL code. Our approach employs a code-to-code augmentation technique to enhance the quality of open-source RTL code datasets. Furthermore, OriGen can rectify syntactic errors through a self-reflection process that leverages compiler feedback.

With the advance of text-to-image (T2I) diffusion models (e.g., Stable Diffusion) and corresponding personalization techniques such as DreamBooth and LoRA, everyone can manifest their imagination into high-quality images at an affordable cost. However, adding motion dynamics to existing high-quality personalized T2Is and enabling them to generate animations remains an open challenge. In this paper, we present AnimateDiff, a practical framework for animating personalized T2I models without requiring model-specific tuning. At the core of our framework is a plug-and-play motion module that can be trained once and seamlessly integrated into any personalized T2Is originating from the same base T2I. Through our proposed training strategy, the motion module effectively learns transferable motion priors from real-world videos. Once trained, the motion module can be inserted into a personalized T2I model to form a personalized animation generator. We further propose MotionLoRA, a lightweight fine-tuning technique for AnimateDiff that enables a pre-trained motion module to adapt to new motion patterns, such as different shot types, at a low training and data collection cost. We evaluate AnimateDiff and MotionLoRA on several public representative personalized T2I models collected from the community. The results demonstrate that our approaches help these models generate temporally smooth animation clips while preserving the visual quality and motion diversity. Codes and pre-trained weights are available at https://github.com/guoyww/AnimateDiff.

Recently video generation has achieved substantial progress with realistic results. Nevertheless, existing AI-generated videos are usually very short clips ("shot-level'') depicting a single scene. To deliver a coherent long video ("story-level''), it is desirable to have creative transition and prediction effects across different clips. This paper presents a short-to-long video diffusion model, SEINE, that focuses on generative transition and prediction. The goal is to generate high-quality long videos with smooth and creative transitions between scenes and varying lengths of shot-level videos. Specifically, we propose a random-mask video diffusion model to automatically generate transitions based on textual descriptions. By providing the images of different scenes as inputs, combined with text-based control, our model generates transition videos that ensure coherence and visual quality. Furthermore, the model can be readily extended to various tasks such as image-to-video animation and autoregressive video prediction. To conduct a comprehensive evaluation of this new generative task, we propose three assessing criteria for smooth and creative transition: temporal consistency, semantic similarity, and video-text semantic alignment. Extensive experiments validate the effectiveness of our approach over existing methods for generative transition and prediction, enabling the creation of story-level long videos.

The extrapolation capability of Large Language Models (LLMs) based on Rotary Position Embedding \citep{su2021roformer} is currently a topic of considerable interest. The mainstream approach to addressing extrapolation with LLMs involves modifying RoPE by replacing 10000, the rotary base of $\theta_n={10000}^{-2n/d}$ in the original RoPE, with a larger value and providing longer fine-tuning text. In this work, we first observe that fine-tuning a RoPE-based LLM with either a smaller or larger base in pre-training context length could significantly enhance its extrapolation performance. After that, we propose \textbf{\textit{Scaling Laws of RoPE-based Extrapolation}}, a unified framework from the periodic perspective, to describe the relationship between the extrapolation performance and base value as well as tuning context length. In this process, we also explain the origin of the RoPE-based extrapolation issue by \textbf{\textit{critical dimension for extrapolation}}. Besides these observations and analyses, we achieve extrapolation up to 1 million context length within only 16K training length on LLaMA2 7B and 13B \citep{touvron2023llama2}.

Despite significant advances in large-scale text-to-image models, achieving hyper-realistic human image generation remains a desirable yet unsolved task. Existing models like Stable Diffusion and DALL·E 2 tend to generate human images with incoherent parts or unnatural poses. To tackle these challenges, our key insight is that human image is inherently structural over multiple granularities, from the coarse-level body skeleton to fine-grained spatial geometry. Therefore, capturing such correlations between the explicit appearance and latent structure in one model is essential to generate coherent and natural human images. To this end, we propose a unified framework, HyperHuman, that generates in-the-wild human images of high realism and diverse layouts. Specifically, 1) we first build a large-scale human-centric dataset, named HumanVerse, which consists of 340M images with comprehensive annotations like human pose, depth, and surface normal. 2) Next, we propose a Latent Structural Diffusion Model that simultaneously denoises the depth and surface normal along with the synthesized RGB image. Our model enforces the joint learning of image appearance, spatial relationship, and geometry in a unified network, where each branch in the model complements to each other with both structural awareness and textural richness. 3) Finally, to further boost the visual quality, we propose a Structure-Guided Refiner to compose the predicted conditions for more detailed generation of higher resolution. Extensive experiments demonstrate that our framework yields the state-of-the-art performance, generating hyper-realistic human images under diverse scenarios.

Human-Scene Interaction (HSI) is a vital component of fields like embodied AI and virtual reality. Despite advancements in motion quality and physical plausibility, two pivotal factors, versatile interaction control and the development of a user-friendly interface, require further exploration before the practical application of HSI. This paper presents a unified HSI framework, UniHSI, which supports unified control of diverse interactions through language commands. The framework defines interaction as ``Chain of Contacts (CoC)", representing steps involving human joint-object part pairs. This concept is inspired by the strong correlation between interaction types and corresponding contact regions. Based on the definition, UniHSI constitutes a Large Language Model (LLM) Planner to translate language prompts into task plans in the form of CoC, and a Unified Controller that turns CoC into uniform task execution. To facilitate training and evaluation, we collect a new dataset named ScenePlan that encompasses thousands of task plans generated by LLMs based on diverse scenarios. Comprehensive experiments demonstrate the effectiveness of our framework in versatile task execution and generalizability to real scanned scenes.

Large language models (LLMs) have demonstrated remarkable performance, and organizations are racing to serve LLMs of varying sizes as endpoints for use-cases like chat, programming and search. However, efficiently serving multiple LLMs poses significant challenges for existing approaches due to varying popularity of LLMs. In the paper, we present MuxServe, a flexible spatial-temporal multiplexing system for efficient multiple LLM serving. The key insight behind is to colocate LLMs considering their popularity to multiplex memory resources, and leverage the characteristics of prefill and decoding phases to separate and flexibly colocate them to multiplex computation resources. MuxServe formally formulates the multiplexing problem, and proposes a novel placement algorithm and adaptive batch scheduling strategy to identify optimal colocations and maximize utilization. MuxServe designs a unified resource manager to enable flexible and efficient multiplexing. Evaluation results show that MuxServe can achieves up to $1.8\times$ higher throughput or processes $2.9\times$ more requests within $99\%$ SLO attainment. The code is available at: https://github.com/hao-ai-lab/MuxServe.

The success of AI assistants based on Language Models (LLMs) hinges on Reinforcement Learning from Human Feedback (RLHF) to comprehend and align with user intentions. However, traditional alignment algorithms, such as PPO, are hampered by complex annotation and training requirements. This reliance limits the applicability of RLHF and hinders the development of professional assistants tailored to diverse human preferences. In this work, we introduce Linear Alignment, a novel algorithm that aligns language models with human preferences in one single inference step, eliminating the reliance on data annotation and model training. Linear alignment incorporates a new parameterization for policy optimization under divergence constraints, which enables the extraction of optimal policy in a closed-form manner and facilitates the direct estimation of the aligned response. Extensive experiments on both general and personalized preference datasets demonstrate that linear alignment significantly enhances the performance and efficiency of LLM alignment across diverse scenarios.

Despite the impressive performance of Reinforcement Learning (RL), the black-box neural network backbone hinders users from trusting and deploying trained agents in real-world applications where safety is crucial. In order to make agents more trustworthy and controllable, for a given RL-trained policy, we propose to enhance its interpretability and make it human-controllable without retraining. We accomplish this goal by following a 3-step pipeline: 1) We interpret neurons by analyzing the causal effect of neurons on the kinematic attributes; To help agents unlock novel skills and enable human to assist agents in accomplishing tasks, 2) we locate the X-neuron, the optimal neuron that is capable of evoking the desired behavior; 3) and edit its activation values to achieve the precise control. We evaluate our method on various RL tasks ranging from autonomous driving to robot locomotion, and the results display that our approach outperforms previous work regarding almost all evaluation metrics. Through enhancing interpretability and introducing human control, the agents can improve safety and performance, even in unseen environments and novel tasks. For locomotion robots simply trained to walk forward, our method unlocks diverse controllable behaviors ranging from jump to backflip.

Kexin Huang, Xiangyang Liu, Qianyu Guo, Tianxiang Sun, Jiawei Sun, Yaru Wang, Zeyang Zhou, Yixu Wang, Yan Teng, Xipeng Qiu, Yingchun Wang, Dahua Lin. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Chonghua Wang, Haodong Duan, Songyang Zhang, Dahua Lin, Kai Chen. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Text-conditioned motion synthesis has made remarkable progress with the emergence of diffusion models. However, the majority of these motion diffusion models are primarily designed for a single character and overlook multi-human interactions. In our approach, we strive to explore this problem by synthesizing human motion with interactions for a group of characters of any size in a zero-shot manner. The key aspect of our approach is the adaptation of human-wise interactions as pairs of human joints that can be either in contact or separated by a desired distance. In contrast to existing methods that necessitate training motion generation models on multi-human motion datasets with a fixed number of characters, our approach inherently possesses the flexibility to model human interactions involving an arbitrary number of individuals, thereby transcending the limitations imposed by the training data. We introduce a novel controllable motion generation method, InterControl, to encourage the synthesized motions maintaining the desired distance between joint pairs. It consists of a motion controller and an inverse kinematics guidance module that realistically and accurately aligns the joints of synthesized characters to the desired location. Furthermore, we demonstrate that the distance between joint pairs for human-wise interactions can be generated using an off-the-shelf Large Language Model (LLM). Experimental results highlight the capability of our framework to generate interactions with multiple human characters and its potential to work with off-the-shelf physics-based character simulators. Code is available at https://github.com/zhenzhiwang/intercontrol.

Human image animation involves generating videos from a character photo, allowing user control and unlocking the potential for video and movie production. While recent approaches yield impressive results using high-quality training data, the inaccessibility of these datasets hampers fair and transparent benchmarking. Moreover, these approaches prioritize 2D human motion and overlook the significance of camera motions in videos, leading to limited control and unstable video generation. To demystify the training data, we present HumanVid, the first large-scale high-quality dataset tailored for human image animation, which combines crafted real-world and synthetic data. For the real-world data, we compile a vast collection of real-world videos from the internet. We developed and applied careful filtering rules to ensure video quality, resulting in a curated collection of 20K high-resolution (1080P) human-centric videos. Human and camera motion annotation is accomplished using a 2D pose estimator and a SLAM-based method. To expand our synthetic dataset, we collected 10K 3D avatar assets and leveraged existing assets of body shapes, skin textures and clothings. Notably, we introduce a rule-based camera trajectory generation method, enabling the synthetic pipeline to incorporate diverse and precise camera motion annotation, which can rarely be found in real-world data. To verify the effectiveness of HumanVid, we establish a baseline model named CamAnimate, short for Camera-controllable Human Animation, that considers both human and camera motions as conditions. Through extensive experimentation, we demonstrate that such simple baseline training on our HumanVid achieves state-of-the-art performance in controlling both human pose and camera motions, setting a new benchmark. Demo, data and code could be found in the project website: https://humanvid.github.io/.

Critique ability, i.e., the capability of Large Language Models (LLMs) to identify and rectify flaws in responses, is crucial for their applications in self-improvement and scalable oversight. While numerous studies have been proposed to evaluate critique ability of LLMs, their comprehensiveness and reliability are still limited. To overcome this problem, we introduce CriticEval, a novel benchmark designed to comprehensively and reliably evaluate critique ability of LLMs. Specifically, to ensure the comprehensiveness, CriticEval evaluates critique ability from four dimensions across nine diverse task scenarios. It evaluates both scalar-valued and textual critiques, targeting responses of varying quality. To ensure the reliability, a large number of critiques are annotated to serve as references, enabling GPT-4 to evaluate textual critiques reliably. Extensive evaluations of open-source and closed-source LLMs first validate the reliability of evaluation in CriticEval. Then, experimental results demonstrate the promising potential of open-source LLMs, the effectiveness of critique datasets and several intriguing relationships between the critique ability and some critical factors, including task types, response qualities and critique dimensions.

We present the ShareGPT4Video series, aiming to facilitate the video understanding of large video-language models (LVLMs) and the video generation of text-to-video models (T2VMs) via dense and precise captions. The series comprises: 1) ShareGPT4Video, 40K GPT4V annotated dense captions of videos with various lengths and sources, developed through carefully designed data filtering and annotating strategy. 2) ShareCaptioner-Video, an efficient and capable captioning model for arbitrary videos, with 4.8M high-quality aesthetic videos annotated by it. 3) ShareGPT4Video-8B, a simple yet superb LVLM that reached SOTA performance on three advancing video benchmarks. To achieve this, taking aside the non-scalable costly human annotators, we find using GPT4V to caption video with a naive multi-frame or frame-concatenation input strategy leads to less detailed and sometimes temporal-confused results. We argue the challenge of designing a high-quality video captioning strategy lies in three aspects: 1) Inter-frame precise temporal change understanding. 2) Intra-frame detailed content description. 3) Frame-number scalability for arbitrary-length videos. To this end, we meticulously designed a differential video captioning strategy, which is stable, scalable, and efficient for generating captions for videos with arbitrary resolution, aspect ratios, and length. Based on it, we construct ShareGPT4Video, which contains 40K high-quality videos spanning a wide range of categories, and the resulting captions encompass rich world knowledge, object attributes, camera movements, and crucially, detailed and precise temporal descriptions of events. Based on ShareGPT4Video, we further develop ShareCaptioner-Video, a superior captioner capable of efficiently generating high-quality captions for arbitrary videos. We annotated 4.8M aesthetically appealing videos by it and verified their effectiveness on a 10-second text2video generation task. For video understanding, we verified the effectiveness of ShareGPT4Video on several current LVLM architectures and presented our superb new LVLM ShareGPT4Video-8B. All the models, strategies, and annotations will be open-sourced and we hope this project can serve as a pivotal resource for advancing both the LVLMs and T2VMs community.

To predict future trajectories, the normalizing flow with a standard Gaussian prior suffers from weak diversity. The ineffectiveness comes from the conflict between the fact of asymmetric and multi-modal distribution of likely outcomes and symmetric and single-modal original distribution and supervision losses. Instead, we propose constructing a mixed Gaussian prior for a normalizing flow model for trajectory prediction. The prior is constructed by analyzing the trajectory patterns in the training samples without requiring extra annotations while showing better expressiveness and being multi-modal and asymmetric. Besides diversity, it also provides better controllability for probabilistic trajectory generation. We name our method Mixed Gaussian Flow (MGF). It achieves state-of-the-art performance in the evaluation of both trajectory alignment and diversity on the popular UCY/ETH and SDD datasets. Code is available at https://github.com/mulplue/MGF.

Large vision-language models (LVLMs) have recently achieved rapid progress, sparking numerous studies to evaluate their multi-modal capabilities. However, we dig into current evaluation works and identify two primary issues: 1) Visual content is unnecessary for many samples. The answers can be directly inferred from the questions and options, or the world knowledge embedded in LLMs. This phenomenon is prevalent across current benchmarks. For instance, GeminiPro achieves 42.7% on the MMMU benchmark without any visual input, and outperforms the random choice baseline across six benchmarks near 24% on average. 2) Unintentional data leakage exists in LLM and LVLM training. LLM and LVLM could still answer some visual-necessary questions without visual content, indicating the memorizing of these samples within large-scale training data. For example, Sphinx-X-MoE gets 43.6% on MMMU without accessing images, surpassing its LLM backbone with 17.9%. Both problems lead to misjudgments of actual multi-modal gains and potentially misguide the study of LVLM. To this end, we present MMStar, an elite vision-indispensable multi-modal benchmark comprising 1,500 samples meticulously selected by humans. MMStar benchmarks 6 core capabilities and 18 detailed axes, aiming to evaluate LVLMs' multi-modal capacities with carefully balanced and purified samples. These samples are first roughly selected from current benchmarks with an automated pipeline, human review is then involved to ensure each curated sample exhibits visual dependency, minimal data leakage, and requires advanced multi-modal capabilities. Moreover, two metrics are developed to measure data leakage and actual performance gain in multi-modal training. We evaluate 16 leading LVLMs on MMStar to assess their multi-modal capabilities, and on 7 benchmarks with the proposed metrics to investigate their data leakage and actual multi-modal gain.

The Large Vision-Language Model (LVLM) field has seen significant advancements, yet its progression has been hindered by challenges in comprehending fine-grained visual content due to limited resolution. Recent efforts have aimed to enhance the high-resolution understanding capabilities of LVLMs, yet they remain capped at approximately 1500 $\times$ 1500 pixels and constrained to a relatively narrow resolution range. This paper represents InternLM-XComposer2-4KHD, a groundbreaking exploration into elevating LVLM resolution capabilities up to 4K HD (3840 × 1600) and beyond. Concurrently, considering the ultra-high resolution may not be necessary in all scenarios, it supports a wide range of diverse resolutions from 336 pixels to 4K standard, significantly broadening its scope of applicability. Specifically, this research advances the patch division paradigm by introducing a novel extension: dynamic resolution with automatic patch configuration. It maintains the training image aspect ratios while automatically varying patch counts and configuring layouts based on a pre-trained Vision Transformer (ViT) (336 $\times$ 336), leading to dynamic training resolution from 336 pixels to 4K standard. Our research demonstrates that scaling training resolution up to 4K HD leads to consistent performance enhancements without hitting the ceiling of potential improvements. InternLM-XComposer2-4KHD shows superb capability that matches or even surpasses GPT-4V and Gemini Pro in 10 of the 16 benchmarks.

Physically realistic materials are pivotal in augmenting the realism of 3D assets across various applications and lighting conditions. However, existing 3D assets and generative models often lack authentic material properties. Manual assignment of materials using graphic software is a tedious and time-consuming task. In this paper, we exploit advancements in Multimodal Large Language Models (MLLMs), particularly GPT-4V, to present a novel approach, Make-it-Real: 1) We demonstrate that GPT-4V can effectively recognize and describe materials, allowing the construction of a detailed material library. 2) Utilizing a combination of visual cues and hierarchical text prompts, GPT-4V precisely identifies and aligns materials with the corresponding components of 3D objects. 3) The correctly matched materials are then meticulously applied as reference for the new SVBRDF material generation according to the original albedo map, significantly enhancing their visual authenticity. Make-it-Real offers a streamlined integration into the 3D content creation workflow, showcasing its utility as an essential tool for developers of 3D assets.

The advent of large vision-language models (LVLMs) has spurred research into their applications in multi-modal contexts, particularly in video understanding. Traditional VideoQA benchmarks, despite providing quantitative metrics, often fail to encompass the full spectrum of video content and inadequately assess models' temporal comprehension. To address these limitations, we introduce MMBench-Video, a quantitative benchmark designed to rigorously evaluate LVLMs' proficiency in video understanding. MMBench-Video incorporates lengthy videos from YouTube and employs free-form questions, mirroring practical use cases. The benchmark is meticulously crafted to probe the models' temporal reasoning skills, with all questions human-annotated according to a carefully constructed ability taxonomy.We employ GPT-4 for automated assessment, demonstrating superior accuracy and robustness over earlier LLM-based evaluations. Utilizing MMBench-Video, we have conducted comprehensive evaluations that include both proprietary and open-source LVLMs for images and videos. MMBench-Video stands as a valuable resource for the research community, facilitating improved evaluation of LVLMs and catalyzing progress in the field of video understanding.

Large language models (LLMs) exhibit hallucinations in long-form question-answering tasks across various domains and wide applications. Current hallucination detection and mitigation datasets are limited in domain and size, which struggle to scale due to prohibitive labor costs and insufficient reliability of existing hallucination annotators. To facilitate the scalable oversight of LLM hallucinations, this paper introduces an iterative self-training framework that simultaneously and progressively scales up the annotation dataset and improves the accuracy of the annotator. Based on the Expectation Maximization algorithm, in each iteration, the framework first applies an automatic hallucination annotation pipeline for a scaled dataset and then trains a more accurate annotator on the dataset. This new annotator is adopted in the annotation pipeline for the next iteration. Extensive experimental results demonstrate that the finally obtained hallucination annotator with only 7B parameters surpasses GPT-4 and obtains new state-of-the-art hallucination detection results on HaluEval and HalluQA by zero-shot inference. Such an annotator can not only evaluate the hallucination levels of various LLMs on the large-scale dataset but also help to mitigate the hallucination of LLMs generations, with the Natural Language Inference metric increasing from 25% to 37% on HaluEval.

The evolution of Artificial Intelligence (AI) has been significantly accelerated by advancements in Large Language Models (LLMs) and Large Multimodal Models (LMMs), gradually showcasing potential cognitive reasoning abilities in problem-solving and scientific discovery (i.e., AI4Science) once exclusive to human intellect. To comprehensively evaluate current models' performance in cognitive reasoning abilities, we introduce OlympicArena, which includes 11,163 bilingual problems across both text-only and interleaved text-image modalities. These challenges encompass a wide range of disciplines spanning seven fields and 62 international Olympic competitions, rigorously examined for data leakage. We argue that the challenges in Olympic competition problems are ideal for evaluating AI's cognitive reasoning due to their complexity and interdisciplinary nature, which are essential for tackling complex scientific challenges and facilitating discoveries. Beyond evaluating performance across various disciplines using answer-only criteria, we conduct detailed experiments and analyses from multiple perspectives. We delve into the models' cognitive reasoning abilities, their performance across different modalities, and their outcomes in process-level evaluations, which are vital for tasks requiring complex reasoning with lengthy solutions. Our extensive evaluations reveal that even advanced models like GPT-4o only achieve a 39.97\% overall accuracy (28.67\% for mathematics and 29.71\% for physics), illustrating current AI limitations in complex reasoning and multimodal integration. Through the OlympicArena, we aim to advance AI towards superintelligence, equipping it to address more complex challenges in science and beyond. We also provide a comprehensive set of resources to support AI research, including a benchmark dataset, an open-source annotation platform, a detailed evaluation tool, and a leaderboard with automatic submission features.

Generating natural and meaningful responses to communicate with multi-modal human inputs is a fundamental capability of Large Vision-Language Models (LVLMs). While current open-source LVLMs demonstrate promising performance in simplified scenarios such as single-turn single-image input, they fall short in real-world conversation scenarios such as following instructions in a long context history with multi-turn and multi-images. Existing LVLM benchmarks primarily focus on single-choice questions or short-form responses, which do not adequately assess the capabilities of LVLMs in real-world human-AI interaction applications. Therefore, we introduce MMDU, a comprehensive benchmark, and MMDU-45k, a large-scale instruction tuning dataset, designed to evaluate and improve LVLMs' abilities in multi-turn and multi-image conversations. We employ the clustering algorithm to find the relevant images and textual descriptions from the open-source Wikipedia and construct the question-answer pairs by human annotators with the assistance of the GPT-4o model.MMDU has a maximum of 18k image+text tokens, 20 images, and 27 turns, which is at least 5x longer than previous benchmarks and poses challenges to current LVLMs. Our in-depth analysis of 15 representative LVLMs using MMDU reveals that open-source LVLMs lag behind closed-source counterparts due to limited conversational instruction tuning data.We demonstrate that fine-tuning open-source LVLMs on MMDU-45k significantly address this gap, generating longer and more accurate conversations, and improving scores on MMDU and existing benchmarks (MMStar: +1.1%, MathVista: +1.5%, ChartQA: +1.2%). Our contributions pave the way for bridging the gap between current LVLM models and real-world application demands. The links to MMDU, and MMDU-45k are available in the supplementary material.

With the emergence of LLMs and their integration with other data modalities, multi-modal 3D perception attracts more attention due to its connectivity to the physical world and makes rapid progress. However, limited by existing datasets, previous works mainly focus on understanding object properties or inter-object spatial relationships in a 3D scene. To tackle this problem, this paper builds the first largest ever multi-modal 3D scene dataset and benchmark with hierarchical grounded language annotations, MMScan. It is constructed based on a top-down logic, from region to object level, from a single target to inter-target relationships, covering holistic aspects of spatial and attribute understanding. The overall pipeline incorporates powerful VLMs via carefully designed prompts to initialize the annotations efficiently and further involve humans' correction in the loop to ensure the annotations are natural, correct, and comprehensive. Built upon existing 3D scanning data, the resulting multi-modal 3D dataset encompasses 1.4M meta-annotated captions on 109k objects and 7.7k regions as well as over 3.04M diverse samples for 3D visual grounding and question-answering benchmarks. We evaluate representative baselines on our benchmarks, analyze their capabilities in different aspects, and showcase the key problems to be addressed in the future. Furthermore, we use this high-quality dataset to train state-of-the-art 3D visual grounding and LLMs and obtain remarkable performance improvement both on existing benchmarks and in-the-wild evaluation.

This paper presents VideoStreaming, an advanced vision-language large model (VLLM) for video understanding, that capably understands arbitrary-length video with a constant number of video tokens streamingly encoded and adaptively selected. The challenge of video understanding in the vision language area mainly lies in the significant computational burden caused by the great number of tokens extracted from long videos. Previous works rely on sparse sampling or frame compression to reduce tokens. However, such approaches either disregard temporal information in a long time span or sacrifice spatial details, resulting in flawed compression. To address these limitations, our VideoStreaming has two core designs: Memory-Propagated Streaming Encoding and Adaptive Memory Selection. The Memory-Propagated Streaming Encoding architecture segments long videos into short clips and sequentially encodes each clip with a propagated memory. In each iteration, we utilize the encoded results of the preceding clip as historical memory, which is integrated with the current clip to distill a condensed representation that encapsulates the video content up to the current timestamp. This method not only incorporates long-term temporal dynamics into the streaming encoding process but also yields a fixed-length memory as a global representation for arbitrarily long videos. After the encoding process, the Adaptive Memory Selection strategy selects a constant number of question-related memories from all the historical memories, and feeds them into the LLM to generate informative responses. The question-related selection reduces redundancy within the memories, enabling efficient and precise video understanding. Meanwhile, the disentangled video extraction and reasoning design allows the LLM to answer different questions about a video by directly selecting corresponding memories, without the need to encode the whole video for each question. Through extensive experiments, our model achieves superior performance and higher efficiency on long video benchmarks, showcasing precise temporal comprehension for detailed question answering.

Vision Language Models (VLMs) demonstrate remarkable proficiency in addressing a wide array of visual questions, which requires strong perception and reasoning faculties. Assessing these two competencies independently is crucial for model refinement, despite the inherent difficulty due to the intertwined nature of seeing and reasoning in existing VLMs. To tackle this issue, we present Prism, an innovative framework designed to disentangle the perception and reasoning processes involved in visual question solving. Prism comprises two distinct stages: a perception stage that utilizes a VLM to extract and articulate visual information in textual form, and a reasoning stage that formulates responses based on the extracted visual information using a Large Language Model (LLM). This modular design enables the systematic comparison and assessment of both proprietary and open-source VLM for their perception and reasoning strengths. Our analytical framework provides several valuable insights, underscoring Prism's potential as a cost-effective solution for vision-language tasks. By combining a streamlined VLM focused on perception with a powerful LLM tailored for reasoning, Prism achieves superior results in general vision-language tasks while substantially cutting down on training and operational expenses. Quantitative evaluations show that Prism, when configured with a vanilla 2B LLaVA and freely accessible GPT-3.5, delivers performance on par with VLMs $10 \times$ larger on the rigorous multimodal benchmark MMStar.

Open-source Large Language Models (LLMs) and their specialized variants, particularly Code LLMs, have recently delivered impressive performance. However, previous Code LLMs are typically fine-tuned on single-source data with limited quality and diversity, which may insufficiently elicit the potential of pre-trained Code LLMs. In this paper, we present AlchemistCoder, a series of Code LLMs with enhanced code generation and generalization capabilities fine-tuned on multi-source data. To achieve this, we pioneer to unveil inherent conflicts among the various styles and qualities in multi-source code corpora and introduce data-specific prompts with hindsight relabeling, termed AlchemistPrompts, to harmonize different data sources and instruction-response pairs. Additionally, we propose incorporating the data construction process into the fine-tuning data as code comprehension tasks, including instruction evolution, data filtering, and code review. Extensive experiments demonstrate that AlchemistCoder holds a clear lead among all models of the same size (6.7B/7B) and rivals or even surpasses larger models (15B/33B/70B), showcasing the efficacy of our method in refining instruction-following capabilities and advancing the boundaries of code intelligence. Source code and models are available at https://github.com/InternLM/AlchemistCoder.

Recent advances in text-to-image generation have enabled the creation of high-quality images with diverse applications. However, accurately describing desired visual attributes can be challenging, especially for non-experts in art and photography. An intuitive solution involves adopting favorable attributes from source images. Current methods attempt to distill identity and style from source images. However, "style" is a broad concept that includes texture, color, and artistic elements, but does not cover other important attributes like lighting and dynamics. Additionally, a simplified "style" adaptation prevents combining multiple attributes from different sources into one generated image. In this work, we formulate a more effective approach to decompose the aesthetics of a picture into specific visual attributes, letting users apply characteristics like lighting, texture, and dynamics from different images. To achieve this goal, we constructed the first fine-grained visual attributes dataset (FiVA) to the best of our knowledge. This FiVA dataset features a well-organized taxonomy for visual attributes and includes 1 M high-quality generated images with visual attribute annotations. Leveraging this dataset, we propose a fine-grained visual attributes adaptation framework (FiVA-Adapter) , which decouples and adapts visual attributes from one or more source images into a generated one. This approach enhances user-friendly customization, allowing users to selectively apply desired attributes to create images that meet their unique preferences and specific content requirements.

Large language models have demonstrated impressive capabilities across various natural language processing tasks, especially in solving mathematical problems. However, large language models are not good at math theorem proving using formal languages like Lean. A significant challenge in this area is the scarcity of training data available in these formal languages. To address this issue, we propose a novel pipeline that iteratively generates and filters synthetic data to translate natural language mathematical problems into Lean 4 statements, and vice versa. Our results indicate that the synthetic data pipeline can provide useful training data and improve the performance of LLMs in translating and understanding complex mathematical problems and proofs. Our final dataset contains about 57K formal-informal question pairs along with searched proof from the math contest forum and 21 new IMO questions. We open-source our code at \url{https://github.com/InternLM/InternLM-Math} and our data at \url{https://huggingface.co/datasets/InternLM/Lean-Workbook}.

Learning powerful representations in bird's-eye-view (BEV) for perception tasks is trending and drawing extensive attention both from industry and academia. Conventional approaches for most autonomous driving algorithms perform detection, segmentation, tracking, etc., in a front or perspective view. As sensor configurations get more complex, integrating multi-source information from different sensors and representing features in a unified view come of vital importance. BEV perception inherits several advantages, as representing surrounding scenes in BEV is intuitive and fusion-friendly; and representing objects in BEV is most desirable for subsequent modules as in planning and/or control. The core problems for BEV perception lie in (a) how to reconstruct the lost 3D information via view transformation from perspective view to BEV; (b) how to acquire ground truth annotations in BEV grid; (c) how to formulate the pipeline to incorporate features from different sources and views; and (d) how to adapt and generalize algorithms as sensor configurations vary across different scenarios. In this survey, we review the most recent works on BEV perception and provide an in-depth analysis of different solutions. Moreover, several systematic designs of BEV approach from the industry are depicted as well. Furthermore, we introduce a full suite of practical guidebook to improve the performance of BEV perception tasks, including camera, LiDAR and fusion inputs. At last, we point out the future research directions in this area. We hope this report will shed some light on the community and encourage more research effort on BEV perception.

3D building reconstruction from monocular remote sensing imagery is an important research problem that has been extensively studied for several decades. Although monocular remote sensing imagery is a more economic data source compared with the LiDAR data and multi-view imagery, its limited information results in great challenges and restricts the performance of existing monocular reconstruction methods. Moreover, the expensive cost and the limited quantity of 3D annotations also restrict the application scenes of existing methods, which are mostly based on fully-supervised learning. In our previous work, we have proposed MTBR-Net, a monocular building reconstruction method that consists of a fully-supervised multi-task network and a post-processing module for optimizing the reconstruction results. In this work, we further propose WS-MTBR-Net, a weakly-supervised building reconstruction network that uses fewer 3D annotations and achieves better performance in an end-to-end manner. Specifically, our WS-MTBR-Net fully leverages the relation between different components of a 3D building instance and the property of off-nadir images to improve the footprint segmentation boundary, based on six modified tasks and a new network structure with an improved feature warping module to support weakly-supervised learning. We also design a new training strategy via a hybrid loss function that enables utilizing the training samples with different annotation levels, i.e., complete 3D annotations, 2D footprint annotations, and image-level angle annotations. Results on BONAI Shanghai and Xi’an test datasets demonstrate that our method achieves competitive performance when using 50% fewer 3D-annotated samples, and improves the footprint segmentation F1-score by around 4% compared with current state-of-the-art.

DNN models are becoming increasingly larger to achieve unprecedented accuracy, and the accompanying increased computation and memory requirements necessitate the employment of massive clusters and elaborate parallelization strategies to accelerate DNN training. In order to better optimize the performance and analyze the cost, it is indispensable to model the training throughput of distributed DNN training. However, complex parallelization strategies and the resulting complex runtime behaviors make it challenging to construct an accurate performance model. In this article, we present Proteus, the first standalone simulator to model the performance of complex parallelization strategies through simulation execution. Proteus first models complex parallelization strategies with a unified representation namedStrategy Tree. Then, it compiles the strategy tree into a distributed execution graph and simulates the complex runtime behaviors,comp-comm overlapandbandwidth sharing, with aHierarchicalTopo-AwareExecutor (HTAE). We finally evaluate Proteus across a wide variety of DNNs on three hardware configurations. Experimental results show that Proteus achieves 3.0% average prediction error and preserves order for training throughput of various parallelization strategies. Compared to state-of-the-art approaches, Proteus reduces prediction error by up to 133.8%.

Knowledge Distillation (KD) aims at distilling the knowledge from the large teacher model to a lightweight student model. Mainstream KD methods can be divided into two categories, logit distillation, and feature distillation. The former is easy to implement, but inferior in performance, while the latter is not applicable to some practical circumstances due to concerns such as privacy and safety. Towards this dilemma, in this paper, we explore a stronger logit distillation method via making better utilization of logit outputs. Concretely, we propose a simple yet effective approach to logit distillation via multi-level prediction alignment. Through this framework, the prediction alignment is not only conducted at the instance level, but also at the batch and class level, through which the student model learns instance prediction, input correlation, and category correlation simultaneously. In addition, a prediction augmentation mechanism based on model calibration further boosts the performance. Extensive experiment results validate that our method enjoys consistently higher performance than previous logit distillation methods, and even reaches competitive performance with mainstream feature distillation methods. Code is available at https://github.com/Jin-Ying/Multi-Level-Logit-Distillation.

This paper presents OmniCity, a new dataset for omnipotent city understanding from multi-level and multi-view images. More precisely, OmniCity contains multi-view satellite images as well as street-level panorama and mono-view images, constituting over 100K pixel-wise annotated images that are well-aligned and collected from 25K geo-locations in New York City. To alleviate the substantial pixel-wise annotation efforts, we propose an efficient street-viewimage annotation pipeline that leverages the existing label maps of satellite view and the transformation relations between different views (satellite, panorama, and mono-view). With the new OmniCity dataset, we provide benchmarks for a variety of tasks including building footprint extraction, height estimation, and building plane/instance/fine-grained segmentation. Compared with existing multi-level and multi-view benchmarks, OmniCity contains a larger number of images with richer annotation types and more views, provides more benchmark results of state-of-the-art models, and introduces a new task for fine-grained building instance segmentation on street-level panorama images. Moreover, OmniCity provides new problem settings for existing tasks, such as cross-view image matching, synthesis, segmentation, detection, etc., and facilitates the developing of new methods for large-scale city understanding, reconstruction, and simulation. The OmniCity dataset as well as the benchmarks will be released at https://city-super.github.io/mnicity/.

Mesh generation is of great value in various applications involving computer graphics and virtual content, yet designing generative models for meshes is challenging due to their irregular data structure and inconsistent topology of meshes in the same category. In this work, we design a novel sparse latent point diffusion model for mesh generation. Our key insight is to regard point clouds as an intermediate representation of meshes, and model the distribution of point clouds instead. While meshes can be generated from point clouds via techniques like Shape as Points (SAP), the challenges of directly generating meshes can be effectively avoided. To boost the efficiency and controllability of our mesh generation method, we propose to further encode point clouds to a set of sparse latent points with pointwise semantic meaningful features, where two DDPMs are trained in the space of sparse latent points to respectively model the distribution of the latent point positions and features at these latent points. We find that sampling in this latent space is faster than directly sampling dense point clouds. Moreover, the sparse latent points also enable us to explicitly control both the overall structures and local details of the generated meshes. Extensive experiments are conducted on the ShapeNet dataset, where our proposed sparse latent point diffusion model achieves superior performance in terms of generation quality and controllability when compared to existing methods. Project page, code and appendix: https://slide-3d.github.io.

This paper studies how to keep a vision backbone effective while removing token mixers in its basic building blocks. Token mixers, as self-attention for vision transformers (ViTs), are intended to perform information communication between different spatial tokens but suffer from considerable computational cost and latency. However, directly removing them will lead to an incomplete model structure prior, and thus brings a significant accuracy drop. To this end, we first develop an RepIdentityFormer base on the re-parameterizing idea, to study the token mixer free model architecture. And we then explore the improved learning paradigm to break the limitation of simple token mixer free backbone, and summarize the empirical practice into 5 guidelines. Equipped with the proposed optimization strategy, we are able to build an extremely simple vision backbone with encouraging performance, while enjoying the high efficiency during inference. Extensive experiments and ablative analysis also demonstrate that the inductive bias of network architecture, can be incorporated into simple network structure with appropriate optimization strategy. We hope this work can serve as a starting point for the exploration of optimization-driven efficient network design.

Recent advances in modeling 3D objects mostly rely on synthetic datasets due to the lack of large-scale real-scanned 3D databases. To facilitate the development of 3D perception, reconstruction, and generation in the real world, we propose OmniObject3D, a large vocabulary 3D object dataset with massive high-quality real-scanned 3D objects. OmniObject3D has several appealing properties: 1) Large Vocabulary: It comprises 6,000 scanned objects in 190 daily categories, sharing common classes with popular 2D datasets (e.g., ImageNet and LVIS), benefiting the pursuit of generalizable 3D representations. 2) Rich Annotations: Each 3D object is captured with both 2D and 3D sensors, providing textured meshes, point clouds, multi-view rendered images, and multiple real-captured videos. 3) Realistic Scans: The professional scanners support high-quality object scans with precise shapes and realistic appearances. With the vast exploration space offered by OmniObject3D, we carefully set up four evaluation tracks: a) robust 3D perception, b) novel-view synthesis, c) neural surface reconstruction, and d) 3D object generation. Extensive studies are performed on these four benchmarks, revealing new observations, challenges, and opportunities for future research in realistic 3D vision.

This paper introduces the Masked Voxel Jigsaw and Reconstruction (MV-JAR) method for LiDAR-based self-supervised pre-training and a carefully designed data-efficient 3D object detection benchmark on the Waymo dataset. Inspired by the scene-voxel-point hierarchy in downstream 3D object detectors, we design masking and reconstruction strategies accounting for voxel distributions in the scene and local point distributions within the voxel. We employ a Reversed-Furthest-Voxel-Sampling strategy to address the uneven distribution of LiDAR points and propose MV-JAR, which combines two techniques for modeling the aforementioned distributions, resulting in superior performance. Our experiments reveal limitations in previous data-efficient experiments, which uniformly sample fine-tuning splits with varying data proportions from each LiDAR sequence, leading to similar data diversity across splits. To address this, we propose a new benchmark that samples scene sequences for diverse fine-tuning splits, ensuring adequate model convergence and providing a more accurate evaluation of pre-training methods. Experiments on our Waymo benchmark and the KITTI dataset demonstrate that MV-JAR consistently and significantly improves 3D detection performance across various data scales, achieving up to a 6.3% increase in mAPH compared to training from scratch. Codes and the benchmark are available at https://github.com/SmartBot-PJLab/MV-JAR.

Purely MLP-based neural radiance fields (NeRF-based methods) often suffer from underfitting with blurred renderings on large-scale scenes due to limited model capacity. Recent approaches propose to geographically divide the scene and adopt multiple sub-NeRFs to model each region individually, leading to linear scale-up in training costs and the number of sub-NeRFs as the scene expands. An alternative solution is to use a feature grid representation, which is computationally efficient and can naturally scale to a large scene with increased grid resolutions. However, the feature grid tends to be less constrained and often reaches suboptimal solutions, producing noisy artifacts in renderings, especially in regions with complex geometry and texture. In this work, we present a new framework that realizes high-fidelity rendering on large urban scenes while being computationally efficient. We propose to use a compact multi-resolution ground feature plane representation to coarsely capture the scene, and complement it with positional encoding inputs through another NeRF branch for rendering in a joint learning fashion. We show that such an integration can utilize the advantages of two alternative solutions: a light-weighted NeRF is sufficient, under the guidance of the feature grid representation, to render photorealistic novel views with fine details; and the jointly optimized ground feature planes, can meanwhile gain further refinements, forming a more accurate and compact feature space and output much more natural rendering results.

Machine learning models with various tensor operators are becoming ubiquitous in recent years. There are two types of operators in machine learning: compute-intensive operators (e.g., GEMM and convolution) and memory-intensive operators (e.g., ReLU and softmax). In emerging machine learning models, compute-intensive operators are usually organized in a chain structure. With the continual specialization of hardware, the gap between computing performance and memory bandwidth has become more prominent. Consequently, the implementations of many compute-intensive operator chains are bounded by memory bandwidth, and generating fused kernels to improve locality for these compute-intensive operators becomes necessary. But in existing machine learning compilers, there lack both precise analysis and efficient optimization for compute-intensive operator chains on different accelerators. As a result, they usually produce sub-optimal performance for these operator chains.In this paper, we propose Chimera, an optimizing framework that can efficiently improve the locality of compute-intensive operator chains on different hardware accelerators. In Chimera, each compute-intensive operator is composed of a series of computation blocks. To generate efficient fused kernels for the operator chains, optimizations for both inter-block and intra-block are required. For inter-block optimization, Chimera decides the optimized block execution order by minimizing the data movement volume among blocks using an analytical model. For intra-block optimization, Chimera uses unified replaceable micro kernels to apply hardware-specific optimizations for different accelerators. Finally, Chimera generates fused kernels for compute-intensive operator chains. Evaluation of batch GEMM chains and convolution chains on CPU, GPU, and NPU shows that Chimera achieves up to 2.87×, 2.29×, and 2.39× speedups to hand-tuned libraries. Compared to state-of-the-art compilers, the speedups are up to 2.29×, 1.64×, and 1.14× for CPU, GPU, and NPU.

Neural radiance fields (NeRF) and its subsequent variants have led to remarkable progress in neural rendering. While most of recent neural rendering works focus on objects and small-scale scenes, developing neural rendering methods for city-scale scenes is of great potential in many real-world applications. However, this line of research is impeded by the absence of a comprehensive and high-quality dataset, yet collecting such a dataset over real city-scale scenes is costly, sensitive, and technically infeasible. To this end, we build a large-scale, comprehensive, and high-quality synthetic dataset for city-scale neural rendering researches. Leveraging the Unreal Engine 5 City Sample project, we developed a pipeline to easily collect aerial and street city views, accompanied by ground-truth camera poses and a range of additional data modalities. Flexible controls on environmental factors like light, weather, human and car crowd are also available in our pipeline, supporting the need of various tasks covering city-scale neural rendering and beyond. The resulting pilot dataset, MatrixCity, contains 67k aerial images and 452k street images from two city maps of total size 28km2. On top of MatrixCity, a thorough benchmark is also conducted, which not only reveals unique challenges of the task of city-scale neural rendering, but also highlights potential improvements for future works. The dataset and code will be publicly available at the project page: https://city-super.github.io/matrixcity/.

Realistic human-centric rendering plays a key role in both computer vision and computer graphics. Rapid progress has been made in the algorithm aspect over the years, yet existing human-centric rendering datasets and benchmarks are rather impoverished in terms of diversity (e.g., outfit's fabric/material, body's interaction with objects, and motion sequences), which are crucial for rendering effect. Researchers are usually constrained to explore and evaluate a small set of rendering problems on current datasets, while real-world applications require methods to be robust across different scenarios. In this work, we present DNA-Rendering, a large-scale, high-fidelity repository of human performance data for neural actor rendering. DNA-Rendering presents several appealing attributes. First, our dataset contains over 1500 human subjects, 5000 motion sequences, and 67.5M frames' data volume. Upon the massive collections, we provide human subjects with grand categories of pose actions, body shapes, clothing, accessories, hairdos, and object intersection, which ranges the geometry and appearance variances from everyday life to professional occasions. Second, we provide rich assets for each subject – 2D/3D human body keypoints, foreground masks, SMPLX models, cloth/accessory materials, multi-view images, and videos. These assets boost the current method's accuracy on downstream rendering tasks. Third, we construct a professional multi-view system to capture data, which contains 60 synchronous cameras with max 4096 × 3000 resolution, 15 fps speed, and stern camera calibration steps, ensuring high-quality resources for task training and evaluation.Along with the dataset, we provide a large-scale and quantitative benchmark in full-scale, with multiple tasks to evaluate the existing progress of novel view synthesis, novel pose animation synthesis, and novel identity rendering methods. In this manuscript, we describe our DNA-Rendering effort as a revealing of new observations, challenges, and future directions to human-centric rendering. The dataset, code, and benchmarks will be publicly available at https://dna-rendering.github.io/.

There has been significant progress in Masked Image Modeling (MIM). Existing MIM methods can be broadly categorized into two groups based on the reconstruction target: pixel-based and tokenizer-based approaches. The former offers a simpler pipeline and lower computational cost, but it is known to be biased toward high-frequency details. In this paper, we provide a set of empirical studies to confirm this limitation of pixel-based MIM and propose a new method that explicitly utilizes low-level features from shallow layers to aid pixel reconstruction. By incorporating this design into our base method, MAE, we reduce the wasted modeling capability of pixel-based MIM, improving its convergence and achieving non-trivial improvements across various downstream tasks. To the best of our knowledge, we are the first to systematically investigate multilevel feature fusion for isotropic architectures like the standard Vision Transformer (ViT). Notably, when applied to a smaller model (e.g., ViT-S), our method yields significant performance gains, such as 1.2% on fine-tuning, 2.8% on linear probing, and 2.6% on semantic segmentation. Code and models are available in MMPretrain1.

Self-supervised methods have shown remarkable progress in learning high-level semantics and low-level temporal correspondence. Building on these results, we take one step further and explore the possibility of integrating these two features to enhance object-centric representations. Our preliminary experiments indicate that query slot attention can extract different semantic components from the RGB feature map, while random sampling based slot attention can exploit temporal correspondence cues between frames to assist instance identification. Motivated by this, we propose a novel semantic-aware masked slot attention on top of the fused semantic features and correspondence maps. It comprises two slot attention stages with a set of shared learnable Gaussian distributions. In the first stage, we use the mean vectors as slot initialization to decompose potential semantics and generate semantic segmentation masks through iterative attention. In the second stage, for each semantics, we randomly sample slots from the corresponding Gaussian distribution and perform masked feature aggregation within the semantic area to exploit temporal correspondence patterns for instance identification. We adopt semantic- and instance-level temporal consistency as self-supervision to encourage temporally coherent object-centric representations. Our model effectively identifies multiple object instances with semantic structure, reaching promising results on unsupervised video object discovery. Furthermore, we achieve state-of-the-art performance on dense label propagation tasks, demonstrating the potential for object-centric analysis. The code is released at https://github.com/shvdiwnkozbw/SMTC.

Human driver can easily describe the complex traffic scene by visual system. Such an ability of precise perception is essential for driver’s planning. To achieve this, a geometry-aware representation that quantizes the physical 3D scene into structured grid map with semantic labels per cell, termed as 3D Occupancy, would be desirable. Compared to the form of bounding box, a key insight behind occupancy is that it could capture the fine-grained details of critical obstacles in the scene, and thereby facilitate subsequent tasks. Prior or concurrent literature mainly concentrate on a single scene completion task, where we might argue that the potential of this occupancy representation might obsess broader impact. In this paper, we propose OccNet, a multi-view vision-centric pipeline with a cascade and temporal voxel decoder to reconstruct 3D occupancy. At the core of OccNet is a general occupancy embedding to represent 3D physical world. Such a descriptor could be applied towards a wide span of driving tasks, including detection, segmentation and planning. To validate the effectiveness of this new representation and our proposed algorithm, we propose OpenOcc, the first dense high-quality 3D occupancy benchmark built on top of nuScenes. Empirical experiments show that there are evident performance gain across multiple tasks, e.g., motion planning could witness a collision rate reduction by 15%-58%, demonstrating the superiority of our method.

Simulation has emerged as an indispensable tool for scaling and accelerating the development of self-driving systems. A critical aspect of this is simulating realistic and diverse human behavior and intent. In this work, we propose a holistic framework for learning physically plausible human dynamics from real driving scenarios, narrowing the gap between real and simulated human behavior in safety-critical applications. We show that state-of-the-art methods underperform in driving scenarios where video data is recorded from moving vehicles, and humans are frequently partially or fully occluded. Furthermore, existing methods often disregard the global scene where humans are situated, resulting in various motion artifacts like foot sliding, floating, or ground penetration. To address this challenge, we propose an approach that incorporates physics with a reinforcement learning-based motion controller to learn human dynamics for driving scenarios. Our framework can simulate physically plausible human dynamics that accurately match observed human motions and infill motions for occluded body parts, while improving the physical plausibility of the entire motion sequence. Experiments on the challenging Waymo Open Dataset show that our method outperforms state-of-the-art motion capture approaches significantly in recovering high-quality, physically plausible, and scene-aware human dynamics.

Recent advances in detecting arbitrary objects in the real world are trained and evaluated on object detection datasets with a relatively restricted vocabulary. To facilitate the development of more general visual object detection, we propose V3Det, a vast vocabulary visual detection dataset with precisely annotated bounding boxes on massive images. V3Det has several appealing properties: 1) Vast Vocabulary: It contains bounding boxes of objects from 13,204 categories on real-world images, which is 10 times larger than the existing large vocabulary object detection dataset, e.g., LVIS. 2) Hierarchical Category Organization: The vast vocabulary of V3Det is organized by a hierarchical category tree which annotates the inclusion relationship among categories, encouraging the exploration of category relationships in vast and open vocabulary object detection. 3) Rich Annotations: V3Det comprises precisely annotated objects in 243k images and professional descriptions of each category written by human experts and a powerful chatbot. By offering a vast exploration space, V3Det enables extensive benchmarks on both vast and open vocabulary object detection, leading to new observations, practices, and insights for future research. It has the potential to serve as a cornerstone dataset for developing more general visual perception systems. V3Det is available at https://v3det.openxlab.org.cn/.

Both indoor and outdoor environments are inherently structured and repetitive. Traditional modeling pipelines keep an asset library storing unique object templates, which is both versatile and memory efficient in practice. Inspired by this observation, we propose AssetField, a novel neural scene representation that learns a set of object-aware ground feature planes to represent the scene, where an asset library storing template feature patches can be constructed in an unsupervised manner. Unlike existing methods which require object masks to query spatial points for object editing, our ground feature plane representation offers a natural visualization of the scene in the bird-eye view, allowing a variety of operations (e.g. translation, duplication, deformation) on objects to configure a new scene. With the template feature patches, group editing is enabled for scenes with many recurring items to avoid repetitive work on object individuals. We show that AssetField not only achieves competitive performance for novel-view synthesis but also generates realistic renderings for new scene configurations.

Synthetic data has emerged as a promising source for 3D human research as it offers low-cost access to large-scale human datasets. To advance the diversity and annotation quality of human models, we introduce a new synthetic dataset, SynBody, with three appealing features: 1) a clothed parametric human model that can generate a diverse range of subjects; 2) the layered human representation that naturally offers high-quality 3D annotations to support multiple tasks; 3) a scalable system for producing realistic data to facilitate real-world tasks. The dataset comprises 1.2M images with corresponding accurate 3D annotations, covering 10,000 human body models, 1,187 actions, and various viewpoints. The dataset includes two subsets for human pose and shape estimation as well as human neural rendering. Extensive experiments on SynBody indicate that it substantially enhances both SMPL and SMPL-X estimation. Furthermore, the incorporation of layered annotations offers a valuable training resource for investigating the Human Neural Radiance Fields(NeRF).

Factor model is a fundamental investment tool in quantitative investment, which can be empowered by deep learning to become more flexible and efficient in practical complicated investing situations. However, it is still an open question to build a factor model that can conduct stock prediction in an online and adaptive setting, where the model can adapt itself to match the current market regime identified based on only point-in-time market information. To tackle this problem, we propose the first deep learning based online and adaptive factor model, HireVAE, at the core of which is a hierarchical latent space that embeds the underlying relationship between the market situation and stock-wise latent factors, so that HireVAE can effectively estimate useful latent factors given only historical market information and subsequently predict accurate stock returns. Across four commonly used real stock market benchmarks, the proposed HireVAE demonstrate superior performance in terms of active returns over previous methods, verifying the potential of such online and adaptive factor model.

Synthesizing high-fidelity head avatars is a central problem for computer vision and graphics. While head avatar synthesis algorithms have advanced rapidly, the best ones still face great obstacles in real-world scenarios. One of the vital causes is the inadequate datasets -- 1) current public datasets can only support researchers to explore high-fidelity head avatars in one or two task directions; 2) these datasets usually contain digital head assets with limited data volume, and narrow distribution over different attributes, such as expressions, ages, and accessories. In this paper, we present RenderMe-360, a comprehensive 4D human head dataset to drive advance in head avatar algorithms across different scenarios. It contains massive data assets, with 243+ million complete head frames and over 800k video sequences from 500 different identities captured by multi-view cameras at 30 FPS. It is a large-scale digital library for head avatars with three key attributes: 1) High Fidelity: all subjects are captured in 360 degrees via 60 synchronized, high-resolution 2K cameras. 2) High Diversity: The collected subjects vary from different ages, eras, ethnicities, and cultures, providing abundant materials with distinctive styles in appearance and geometry. Moreover, each subject is asked to perform various dynamic motions, such as expressions and head rotations, which further extend the richness of assets. 3) Rich Annotations: the dataset provides annotations with different granularities: cameras' parameters, background matting, scan, 2D/3D facial landmarks, FLAME fitting, and text description. Based on the dataset, we build a comprehensive benchmark for head avatar research, with 16 state-of-the-art methods performed on five main tasks: novel view synthesis, novel expression synthesis, hair rendering, hair editing, and talking head generation. Our experiments uncover the strengths and flaws of state-of-the-art methods. RenderMe-360 opens the door for future exploration in modern head avatars. All of the data, code, and models will be publicly available at https://renderme-360.github.io/.

3D content creation from a single image is a long-standing yet highly desirable task. Recent advances introduce 2D diffusion priors, yielding reasonable results. However, existing methods are not hyper-realistic enough for post-generation usage, as users cannot view, render and edit the resulting 3D content from a full range. To address these challenges, we introduce HyperDreamer with several key designs and appealing properties: 1) Full-range viewable: 360° mesh modeling with high-resolution textures enables the creation of visually compelling 3D models from a full range of observation points. 2) Full-range renderable: Fine-grained semantic segmentation and data-driven priors are incorporated as guidance to learn reasonable albedo, roughness, and specular properties of the materials, enabling semantic-aware arbitrary material estimation. 3) Full-range editable: For a generated model or their own data, users can interactively select any region via a few clicks and efficiently edit the texture with text-based guidance. Extensive experiments demonstrate the effectiveness of HyperDreamer in modeling region-aware materials with high-resolution textures and enabling user-friendly editing. We believe that HyperDreamer holds promise for advancing 3D content creation and finding applications in various domains.

We present an end-to-end system for the high-fidelity capture, model reconstruction, and real-time rendering of walkable spaces in virtual reality using neural radiance fields. To this end, we designed and built a custom multi-camera rig to densely capture walkable spaces in high fidelity and with multi-view high dynamic range images in unprecedented quality and density. We extend instant neural graphics primitives with a novel perceptual color space for learning accurate HDR appearance, and an efficient mip-mapping mechanism for level-of-detail rendering with anti-aliasing, while carefully optimizing the trade-off between quality and speed. Our multi-GPU renderer enables high-fidelity volume rendering of our neural radiance field model at the full VR resolution of dual 2K × 2K at 36 Hz on our custom demo machine. We demonstrate the quality of our results on our challenging high-fidelity datasets, and compare our method and datasets to existing baselines. We release our dataset on our project website: https://vr-nerf.github.io.

End-to-end scene text spotting has made significant progress due to its intrinsic synergy between text detection and recognition. Previous methods commonly regard manual annotations such as horizontal rectangles, rotated rectangles, quadrangles, and polygons as a prerequisite, which are much more expensive than using single-point. Our new framework, SPTS v2, allows us to train high-performing text-spotting models using a single-point annotation. SPTS v2 reserves the advantage of the auto-regressive Transformer with an Instance Assignment Decoder (IAD) through sequentially predicting the center points of all text instances inside the same predicting sequence, while with a Parallel Recognition Decoder (PRD) for text recognition in parallel, which significantly reduces the requirement of the length of the sequence. These two decoders share the same parameters and are interactively connected with a simple but effective information transmission process to pass the gradient and information. Comprehensive experiments on various existing benchmark datasets demonstrate the SPTS v2 can outperform previous state-of-the-art single-point text spotters with fewer parameters while achieving 19× faster inference speed. Within the context of our SPTS v2 framework, our experiments suggest a potential preference for single-point representation in scene text spotting when compared to other representations. Such an attempt provides a significant opportunity for scene text spotting applications beyond the realms of existing paradigms.

Video unscreen, a technique to extract foreground from given videos, has been playing an important role in today's video production pipeline. Existing systems developed for this purpose which mainly rely on video segmentation or video matting, either suffer from quality deficiencies or requiring tedious manual annotations. In this work, we aim to develop a fully automatic video unscreen framework that is able to obtain high-quality foreground extraction without the need of human intervention in a controlled environment. Inspired by the alpha composition equation, our frame adopts a coarse-to-fine strategy, where the obtained background estimate given an initial mask prediction in turn helps the refinement of the mask. We conducted experiments on two datasets, 1) the Adobe's Synthetic-Composite dataset, and 2) DramaStudio, our newly collected large-scale green screen video matting dataset, exhibiting the controlled environments. The results show that the proposed framework outperforms existing algorithms and commercial software, both quantitatively and qualitatively. We also demonstrate its utility in person replacement in videos, which can further support a variety of video editing applications.

Anime is an abstract art form that is substantially different from the human portrait, leading to a challenging misaligned image translation problem that is beyond the capability of existing methods. This can be boiled down to a highly ambiguous unconstrained translation between two domains. To this end, we design a new anime translation framework by deriving the prior knowledge of a pre-trained StyleGAN model. We introduce disentangled encoders to separately embed structure and appearance information into the same latent code, governed by four tailored losses. Moreover, we develop a FaceBank aggregation method that leverages the generated data of the StyleGAN, anchoring the prediction to produce in-domain animes. To empower our model and promote the research of anime translation, we propose the first anime portrait parsing dataset, Danbooru-Parsing , containing 4,921 densely labeled images across 17 classes. This dataset connects the face semantics with appearances, enabling our new constrained translation setting. We further show the editability of our results, and extend our method to manga images, by generating the first manga parsing pseudo data. Extensive experiments demonstrate the values of our new dataset and method, resulting in the first feasible solution on anime translation.

The ability to synthesize long-term human motion sequences in real-world scenes can facilitate numerous applications. Previous approaches for scene-aware motion synthesis are constrained by pre-defined target objects or positions and thus limit the diversity of human-scene interactions for synthesized motions. In this paper, we focus on the problem of synthesizing diverse scene-aware human motions under the guidance of target action sequences. To achieve this, we first decompose the diversity of scene-aware human motions into three aspects, namely interaction diversity (e.g. sitting on different objects with different poses in the given scenes), path diversity (e.g. moving to the target locations following different paths), and the motion diversity (e.g. having various body movements during moving). Based on this factorized scheme, a hierarchical framework is proposed, with each sub-module responsible for modeling one aspect. We assess the effectiveness of our framework on two challenging datasets for scene-aware human motion synthesis. The experiment results show that the proposed framework remarkably outperforms previous methods in terms of diversity and naturalness.

Human skeleton, as a compact representation of human action, has received increasing attention in recent years. Many skeleton-based action recognition methods adopt GCNs to extract features on top of human skeletons. Despite the positive results shown in these attempts, GCN-based methods are subject to limitations in robustness, interoperability, and scalability. In this work, we propose PoseConv3D, a new approach to skeleton-based action recognition. PoseConv3D relies on a 3D heatmap volume instead of a graph sequence as the base representation of human skeletons. Compared to GCN-based methods, PoseConv3D is more effective in learning spatiotemporal features, more robust against pose estimation noises, and generalizes better in cross-dataset settings. Also, PoseConv3D can handle multiple-person scenarios without additional computation costs. The hierarchical features can be easily integrated with other modalities at early fusion stages, providing a great design space to boost the performance. PoseConv3D achieves the state-of-the-art on five of six standard skeleton-based action recognition benchmarks. Once fused with other modalities, it achieves the state-of-the-art on all eight multi-modality action recognition benchmarks. Code has been made available at: https://github.com/kennymckormick/pyskl.

Recognizing transformation types applied to a video clip (RecogTrans) is a long-established paradigm for selfsupervised video representation learning, which achieves much inferior performance compared to instance discrimination approaches (InstDisc) in recent works. However, based on a thorough comparison of representative Recog-Trans and InstDisc methods, we observe the great potential of RecogTrans on both semantic-related and temporalrelated downstream tasks. Based on hard-label classification, existing RecogTrans approaches suffer from noisy supervision signals in pre-training. To mitigate this problem, we developed TransRank, a unified framework for recognizing Transformations in a Ranking formulation. TransRank provides accurate supervision signals by recognizing transformations relatively, consistently outperforming the classification-based formulation. Meanwhile, the unified framework can be instantiated with an arbitrary set of temporal or spatial transformations, demonstrating good generality. With a ranking-based formulation and several empirical practices, we achieve competitive performance on video retrieval and action recognition. Under the same setting, TransRank surpasses the previous state-of-the-art method [28] by 6.4% on UCF101 and 8.3% on HMDB51 for action recognition (Topl Acc); improves video retrieval on UCF101 by 20.4% (R@1). The promising results validate that RecogTrans is still a worth exploring paradigm for video self-supervised learning. Codes will be released at https://github.com/kennymckormick/TransRank.

End-to-end scene text spotting has attracted great attention in recent years due to the success of excavating the intrinsic synergy of the scene text detection and recognition. However, recent state-of-the-art methods usually incorporate detection and recognition simply by sharing the backbone, which does not directly take advantage of the feature interaction between the two tasks. In this paper, we propose a new end-to-end scene text spotting framework termed SwinTextSpotter. Using a transformer encoder with dynamic head as the detector, we unify the two tasks with a novel Recognition Conversion mechanism to explicitly guide text localization through recognition loss. The straightforward design results in a concise framework that requires neither additional rectification module nor character-level annotation for the arbitrarily-shaped text. Qualitative and quantitative experiments on multi-oriented datasets RoIC13 and ICDAR 2015, arbitrarily-shaped datasets Total-Text and CTW1500, and multi-lingual datasets ReCTS (Chinese) and VinText (Viet-namese) demonstrate SwinTextSpotter significantly outperforms existing methods. Code is available at https://github.com/mxin262/SwinTextSpotter.

Videos incorporate rich semantics as well as redundant information. Seeking a compact yet effective video representation, e.g., sample informative frames from the entire video, is critical to efficient video recognition. There have been works that formulate frame sampling as a sequential decision task by selecting frames one by one according to their importance. In this paper, we present a more efficient framework named OCSampler, which explores such a representation with one short clip. OCSampler designs a new paradigm of learning instance-specific video condensation policies to select frames only in a single step. Rather than picking up frames sequentially like previous methods, we simply process a whole sequence at once. Accordingly, these policies are derived from a light-weighted skim network together with a simple yet effective policy network. Moreover, we extend the proposed method with a frame number budget, enabling the framework to produce correct predictions in high confidence with as few frames as possible. Experiments on various benchmarks demonstrate the effectiveness of OCSampler over previous methods in terms of accuracy and efficiency. Specifically, it achieves 76.9% mAP and 21.7 GFLOPs on ActivityNet with an impressive throughput: 123.9 Video/s on a single TITAN Xp GPU.

In recent years, memory-intensive operations are becoming dominant in efficiency of running novel neural networks. Just-in-time operator fusion on accelerating devices like GPU proves an effective method for optimizing memory-intensive operations, and suits the numerous varying model structures. In particular, we find memory-intensive operations on tensor views are ubiquitous in neural network implementations. Tensors are the de facto representation for numerical data in deep learning areas, while tensor views cover a bunch of sophisticated syntax, which allow various interpretations on the underlying tensor data without memory copy. The support of views in deep learning compilers could greatly enlarge operator fusion scope, and appeal to optimizing novel neural networks. Nevertheless, mainstream solutions in state-of-the-art deep learning compilers exhibit imperfections either in view syntax representations or operator fusion. In this article, we propose EasyView, which enables and schedules tensor views in an end-to-end workflow from neural networks onto devices. Aiming at maximizing memory utilization and reducing data movement, we categorize various view contexts in high-level language, and lower views in accordance with different scenarios. Reference-semantic in terms of views are kept in the lowering from native high-level language features to intermediate representations. Based on the reserved reference-semantics, memory activities related to data dependence of read and write are tracked for further compute and memory optimization. Besides, ample operator fusion is applied to memory-intensive operations with views. In our tests, the proposed work could get average 5.63X, 2.44X, and 4.67X speedup compared with the XLA, JAX, and TorchScript, respectively for hotspot Python functions. In addition, operation fusion with views could bring 8.02% performance improvement in end-to-end neural networks.

We present PYSKL: an open-source toolbox for skeleton-based action recognition based on PyTorch. The toolbox supports a wide variety of skeleton action recognition algorithms, including approaches based on GCN and CNN. In contrast to existing open-source skeleton action recognition projects that include only one or two algorithms, PYSKL implements six different algorithms under a unified framework with both the latest and original good practices to ease the comparison of efficacy and efficiency. We also provide an original GCN-based skeleton action recognition model named ST-GCN++, which achieves competitive recognition performance without any complicated attention schemes, serving as a strong baseline. Meanwhile, PYSKL supports the training and testing of nine skeleton-based action recognition benchmarks and achieves state-of-the-art recognition performance on eight of them. To facilitate future research on skeleton action recognition, we also provide a large number of trained models and detailed benchmark results to give some insights. PYSKL is released at https://github.com/kennymckormick/pyskl and is actively maintained.

Existing scene text spotting (i.e., end-to-end text detection and recognition) methods rely on costly bounding box annotations (e.g., text-line, word-level, or character-level bounding boxes). For the first time, we demonstrate that training scene text spotting models can be achieved with an extremely low-cost annotation of a single-point for each instance. We propose an end-to-end scene text spotting method that tackles scene text spotting as a sequence prediction task. Given an image as input, we formulate the desired detection and recognition results as a sequence of discrete tokens and use an auto-regressive Transformer to predict the sequence. The proposed method is simple yet effective, which can achieve state-of-the-art results on widely used benchmarks. Most significantly, we show that the performance is not very sensitive to the positions of the point annotation, meaning that it can be much easier to be annotated or even be automatically generated than the bounding box that requires precise positions. We believe that such a pioneer attempt indicates a significant opportunity for scene text spotting applications of a much larger scale than previously possible. The code is available at https://github.com/shannanyinxiang/SPTS.

Among numerous videos shared on the web, well-edited ones always attract more attention. However, it is difficult for inexperienced users to make well-edited videos because it requires professional expertise and immense manual labor. To meet the demands for non-experts, we present Transcript-to-Video -- a weakly-supervised framework that uses texts as input to automatically create video sequences from an extensive collection of shots. Specifically, we propose a Content Retrieval Module and a Temporal Coherent Module to learn visual-language representations and model shot sequencing styles, respectively. For fast inference, we introduce an efficient search strategy for real-time video clip sequencing. Quantitative results and user studies demonstrate empirically that the proposed learning framework can retrieve content-relevant shots while creating plausible video sequences in terms of style. Besides, the run-time performance analysis shows that our framework can support real-world applications. Project page: http://www.xiongyu.me/projects/transcript2video/

Semi-Supervised Semantic Segmentation aims at training the segmentation model with limited labeled data and a large amount of unlabeled data. To effectively leverage the unlabeled data, pseudo labeling, along with the teacher-student framework, is widely adopted in semi-supervised semantic segmentation. Though proved to be effective, this paradigm suffers from incorrect pseudo labels which inevitably exist and are taken as auxiliary training data. To alleviate the negative impact of incorrect pseudo labels, we delve into the current Semi-Supervised Semantic Segmentation frameworks. We argue that the unlabeled data with pseudo labels can facilitate the learning of representative features in the feature extractor, but it is unreliable to supervise the mask predictor. Motivated by this consideration, we propose a novel framework, Gentle Teaching Assistant (GTA-Seg) to disentangle the effects of pseudo labels on feature extractor and mask predictor of the student model. Specifically, in addition to the original teacher-student framework, our method introduces a teaching assistant network which directly learns from pseudo labels generated by the teacher network. The gentle teaching assistant (GTA) is coined gentle since it only transfers the beneficial feature representation knowledge in the feature extractor to the student model in an Exponential Moving Average (EMA) manner, protecting the student model from the negative influences caused by unreliable pseudo labels in the mask predictor. The student model is also supervised by reliable labeled data to train an accurate mask predictor, further facilitating feature representation. Extensive experiment results on benchmark datasets validate that our method shows competitive performance against previous methods. We promise to release our code towards reproducibility.

Co-speech gesture is crucial for human-machine interaction and digital entertainment. While previous works mostly map speech audio to human skeletons (e.g., 2D keypoints), directly generating speakers' gestures in the image domain remains unsolved. In this work, we formally define and study this challenging problem of audio-driven co-speech gesture video generation, i.e., using a unified framework to generate speaker image sequence driven by speech audio. Our key insight is that the co-speech gestures can be decomposed into common motion patterns and subtle rhythmic dynamics. To this end, we propose a novel framework, Audio-driveN Gesture vIdeo gEneration (ANGIE), to effectively capture the reusable co-speech gesture patterns as well as fine-grained rhythmic movements. To achieve high-fidelity image sequence generation, we leverage an unsupervised motion representation instead of a structural human body prior (e.g., 2D skeletons). Specifically, 1) we propose a vector quantized motion extractor (VQ-Motion Extractor) to summarize common co-speech gesture patterns from implicit motion representation to codebooks. 2) Moreover, a co-speech gesture GPT with motion refinement (Co-Speech GPT) is devised to complement the subtle prosodic motion details. Extensive experiments demonstrate that our framework renders realistic and vivid co-speech gesture video. Demo video and more resources can be found in: https://alvinliu0.github.io/projects/ANGIE

Learning a good image prior is a long-term goal for image restoration and manipulation. While existing methods like deep image prior (DIP) capture low-level image statistics, there are still gaps toward an image prior that captures rich image semantics including color, spatial coherence, textures, and high-level concepts. This work presents an effective way to exploit the image prior captured by a generative adversarial network (GAN) trained on large-scale natural images. As shown in Fig. 1, the deep generative prior (DGP) provides compelling results to restore missing semantics, e.g., color, patch, resolution, of various degraded images. It also enables diverse image manipulation including random jittering, image morphing, and category transfer. Such highly flexible restoration and manipulation are made possible through relaxing the assumption of existing GAN inversion methods, which tend to fix the generator. Notably, we allow the generator to be fine-tuned on-the-fly in a progressive manner regularized by feature distance obtained by the discriminator in GAN. We show that these easy-to-implement and practical changes help preserve the reconstruction to remain in the manifold of nature images, and thus lead to more precise and faithful reconstruction for real images. Code is available at https://github.com/XingangPan/deep-generative-prior.

Feature reassembly, i.e. feature downsampling and upsampling, is a key operation in a number of modern convolutional network architectures, e.g., residual networks and feature pyramids. Its design is critical for dense prediction tasks such as object detection and semantic/instance segmentation. In this work, we propose unified Content-Aware ReAssembly of FEatures (CARAFE++), a universal, lightweight, and highly effective operator to fulfill this goal. CARAFE++ has several appealing properties: (1) Unlike conventional methods such as pooling and interpolation that only exploit sub-pixel neighborhood, CARAFE++ aggregates contextual information within a large receptive field. (2) Instead of using a fixed kernel for all samples (e.g. convolution and deconvolution), CARAFE++ generates adaptive kernels on-the-fly to enable instance-specific content-aware handling. (3) CARAFE++ introduces little computational overhead and can be readily integrated into modern network architectures. We conduct comprehensive evaluations on standard benchmarks in object detection, instance/semantic segmentation, and image inpainting. CARAFE++ shows consistent and substantial gains on mainstream methods across all the tasks with negligible computational overhead. It shows great potential to serve as a strong building block for modern deep networks.

State-of-the-art methods for driving-scene LiDAR-based perception (including point cloud semantic segmentation, panoptic segmentation and 3D detection, etc.) often project the point clouds to 2D space and then process them via 2D convolution. Although this cooperation shows the competitiveness in the point cloud, it inevitably alters and abandons the 3D topology and geometric relations. A natural remedy is to utilize the 3D voxelization and 3D convolution network. However, we found that in the outdoor point cloud, the improvement obtained in this way is quite limited. An important reason is the property of the outdoor point cloud, namely sparsity and varying density. Motivated by this investigation, we propose a new framework for the outdoor LiDAR segmentation, where cylindrical partition and asymmetrical 3D convolution networks are designed to explore the 3D geometric pattern while maintaining these inherent properties. The proposed model acts as a backbone and the learned features from this model can be used for downstream tasks such as point cloud semantic and panoptic segmentation or 3D detection. In this paper, we benchmark our model on these three tasks. For semantic segmentation, we evaluate the proposed model on several large-scale datasets, i.e., SemanticKITTI, nuScenes and A2D2. Our method achieves the state-of-the-art on the leaderboard of SemanticKITTI (both single-scan and multi-scan challenge), and significantly outperforms existing methods on nuScenes and A2D2 dataset. Furthermore, the proposed 3D framework also shows strong performance and good generalization on LiDAR panoptic segmentation and LiDAR 3D detection.

In film making, shot has a profound influence on how the movie content is delivered and how the audiences are echoed, where different emotions and contents can be delivered through well-designed camera movements or shot editing. Therefore, in pursuit of a high-level understanding of long videos, accurate shot detection from untrimmed videos should be considered as the first and the most fundamental step. Existing approaches address this problem based on the visual difference and content transitions between consecutive frames, while ignoring intrinsic shot attributes, viz., camera movements, scales and viewing angles, which essentially reveals how each shot is created. In this work, we propose a new learning framework (SCTSNet) for shot boundary detection by jointly recognizing the attributes and composition of shots in videos. To facilitate the analysis of shots and the evaluation of shot detection models, we collect a large-scale shot boundary dataset MovieShots2, which contains 15K shots from 282 movie clips. It is richly annotated with the temporal boundary between consecutive shots and its shot attributes, including camera movements, scales and viewing angles, which are the three most distinct shot attributes. Our experiments show that the joint learning framework can significantly boost the boundary detection performance, surpassing the previous scores by a large margin. SCTSNet improves shot boundary detection AP from 0.65 to 0.77, pushing the performance to a new level.

While tremendous advances in visual and auditory realism have been made for virtual and augmented reality (VR/AR), introducing a plausible sense of physicality into the virtual world remains challenging. Closing the gap between real-world physicality and immersive virtual experience requires a closed interaction loop: applying user-exerted physical forces to the virtual environment and generating haptic sensations back to the users. However, existing VR/AR solutions either completely ignore the force inputs from the users or rely on obtrusive sensing devices that compromise user experience. By identifying users' muscle activation patterns while engaging in VR/AR, we design a learning-based neural interface for natural and intuitive force inputs. Specifically, we show that lightweight electromyography sensors, resting non-invasively on users' forearm skin, inform and establish a robust understanding of their complex hand activities. Fuelled by a neural-network-based model, our interface can decode finger-wise forces in real-time with 3.3% mean error, and generalize to new users with little calibration. Through an interactive psychophysical study, we show that human perception of virtual objects' physical properties, such as stiffness, can be significantly enhanced by our interface. We further demonstrate that our interface enables ubiquitous control via finger tapping. Ultimately, we envision our findings to push forward research towards more realistic physicality in future VR/AR.

Video object detection is challenging in the presence of appearance deterioration in certain video frames. Therefore, it is a natural choice to aggregate temporal information from other frames of the same video into the current frame. However, ROI Align, as one of the most core procedures of video detectors, still remains extracting features from a single-frame feature map for proposals, making the extracted ROI features lack temporal information from videos. In this work, considering the features of the same object instance are highly similar among frames in a video, a novel Temporal ROI Align operator is proposed to extract features from other frames feature maps for current frame proposals by utilizing feature similarity. The proposed Temporal ROI Align operator can extract temporal information from the entire video for proposals. We integrate it into single-frame video detectors and other state-of-the-art video detectors, and conduct quantitative experiments to demonstrate that the proposed Temporal ROI Align operator can consistently and significantly boost the performance. Besides, the proposed Temporal ROI Align can also be applied into video instance segmentation.

Building extraction from aerial or satellite images has been an important research issue in remote sensing and computer vision domains for decades. Compared with pixel-wise semantic segmentation models that output raster building segmentation map, polygonal building segmentation approaches produce more realistic building polygons that are in the desirable vector format for practical applications. Despite the substantial efforts over recent years, state-of-the-art polygonal building segmentation methods still suffer from several limitations, e.g., (1) relying on a perfect segmentation map to guarantee the vectorization quality; (2) requiring a complex post-processing procedure; (3) generating inaccurate vertices with a fixed quantity, a wrong sequential order, self-intersections, etc. To tackle the above issues, in this paper, we propose a polygonal building segmentation approach and make the following contributions: (1) We design a multi-task segmentation network for joint semantic and geometric learning via three tasks, i.e., pixel-wise building segmentation, multi-class corner prediction, and edge orientation prediction. (2) We propose a simple but effective vertex generation module for transforming the segmentation contour into high-quality polygon vertices. (3) We further propose a polygon refinement network that automatically moves the polygon vertices into more accurate locations. Results on two popular building segmentation datasets demonstrate that our approach achieves significant improvements for both building instance segmentation (with 2% F1-score gain) and polygon vertex prediction (with 6% F1-score gain) compared with current state-of-the-art methods.

Controversy exists on whether differentiable neural architecture search methods discover wiring topology effectively. To understand how wiring topology evolves, we study the underlying mechanism of several existing differentiable NAS frameworks. Our investigation is motivated by three observed searching patterns of differentiable NAS: 1) they search by growing instead of pruning; 2) wider networks are more preferred than deeper ones; 3) no edges are selected in bi-level optimization. To anatomize these phenomena, we propose a unified view on searching algorithms of existing frameworks, transferring the global optimization to local cost minimization. Based on this reformulation, we conduct empirical and theoretical analyses, revealing implicit biases in the cost’s assignment mechanism and evolution dynamics that cause the observed phenomena. These biases indicate strong discrimination towards certain topologies. To this end, we pose questions that future differentiable methods for neural wiring discovery need to confront, hoping to evoke a discussion and rethinking on how much bias has been enforced implicitly in existing NAS methods.

Recent works have shown that interval bound propagation (IBP) can be used to train verifiably robust neural networks. Reseachers observe an intriguing phenomenon on these IBP trained networks: CROWN, a bounding method based on tight linear relaxation, often gives very loose bounds on these networks. We also observe that most neurons become dead during the IBP training process, which could hurt the representation capability of the network. In this paper, we study the relationship between IBP and CROWN, and prove that CROWN is always tighter than IBP when choosing appropriate bounding lines. We further propose a relaxed version of CROWN, linear bound propagation (LBP), that can be used to verify large networks to obtain lower verified errors than IBP. We also design a new activation function, parameterized ramp function (ParamRamp), which has more diversity of neuron status than ReLU. We conduct extensive experiments on MNIST, CIFAR-10 and Tiny-ImageNet with ParamRamp activation and achieve state-of-the-art verified robustness. Code is available at https://github.com/ZhaoyangLyu/VerifiablyRobustNN.

We revisit human motion synthesis, a task useful in various real-world applications, in this paper. Whereas a number of methods have been developed previously for this task, they are often limited in two aspects: 1) focus on the poses while leaving the location movement behind, and 2) ignore the impact of the environment on the human motion. In this paper, we propose a new framework, with the interaction between the scene and the human motion taken into account. Considering the uncertainty of human motion, we formulate this task as a generative task, whose objective is to generate plausible human motion conditioned on both the scene and the human’s initial position. This framework factorizes the distribution of human motions into a distribution of movement trajectories conditioned on scenes and that of body pose dynamics conditioned on both scenes and trajectories. We further derive a GAN-based learning approach, with discriminators to enforce the compatibility between the human motion and the contextual scene as well as the 3D-to-2D projection constraints. We assess the effectiveness of the proposed method on two challenging datasets, which cover both synthetic and real-world environments.

Instance segmentation has witnessed a remarkable progress on class-balanced benchmarks. However, they fail to perform as accurately in real-world scenarios, where the category distribution of objects naturally comes with a long tail. Instances of head classes dominate a long-tailed dataset and they serve as negative samples of tail categories. The overwhelming gradients of negative samples on tail classes lead to a biased learning process for classifiers. Consequently, objects of tail categories are more likely to be misclassified as backgrounds or head categories. To tackle this problem, we propose Seesaw Loss to dynamically re-balance gradients of positive and negative samples for each category, with two complementary factors, i.e., mitigation factor and compensation factor. The mitigation factor reduces punishments to tail categories w.r.t. the ratio of cumulative training instances between different categories. Meanwhile, the compensation factor increases the penalty of misclassified instances to avoid false positives of tail categories. We conduct extensive experiments on Seesaw Loss with mainstream frameworks and different data sampling strategies. With a simple end-to-end training pipeline, Seesaw Loss obtains significant gains over Cross-Entropy Loss, and achieves state-of-the-art performance on LVIS dataset without bells and whistles. Code is available at https://github.com/open-mmlab/mmdetection.

Adversarial robustness has attracted extensive studies recently by revealing the vulnerability and intrinsic characteristics of deep networks. However, existing works on adversarial robustness mainly focus on balanced datasets, while real-world data usually exhibits a long-tailed distribution. To push adversarial robustness towards more realistic scenarios, in this work we investigate the adversarial vulnerability as well as defense under long-tailed distributions. In particular, we first reveal the negative impacts induced by imbalanced data on both recognition performance and adversarial robustness, uncovering the intrinsic challenges of this problem. We then perform a systematic study on existing long-tailed recognition methods in conjunction with the adversarial training framework. Several valuable observations are obtained: 1) natural accuracy is relatively easy to improve, 2) fake gain of robust accuracy exists under unreliable evaluation, and 3) boundary error limits the promotion of robustness. Inspired by these observations, we propose a clean yet effective framework, RoBal, which consists of two dedicated modules, a scale-invariant classifier and data re-balancing via both margin engineering at training stage and boundary adjustment during inference. Extensive experiments demonstrate the superiority of our approach over other state-of-the-art defense methods. To our best knowledge, we are the first to tackle adversarial robustness under long-tailed distributions, which we believe would be a significant step towards real-world robustness. Our code is available at: https://github.com/wutong16/Adversarial_Long-Tail.

Stereophonic audio, especially binaural audio, plays an essential role in immersive viewing environments. Recent research has explored generating visually guided stereophonic audios supervised by multi-channel audio collections. However, due to the requirement of professional recording devices, existing datasets are limited in scale and variety, which impedes the generalization of supervised methods in real-world scenarios. In this work, we propose PseudoBinaural, an effective pipeline that is free of binaural recordings. The key insight is to carefully build pseudo visual-stereo pairs with mono data for training. Specifically, we leverage spherical harmonic decomposition and head-related impulse response (HRIR) to identify the relationship between spatial locations and received binaural audios. Then in the visual modality, corresponding visual cues of the mono data are manually placed at sound source positions to form the pairs. Compared to fully-supervised paradigms, our binaural-recording-free pipeline shows great stability in cross-dataset evaluation and achieves comparable performance under subjective preference. Moreover, combined with binaural recordings, our method is able to further boost the performance of binaural audio generation under supervised settings1.

3D building reconstruction from monocular remote sensing imagery is an important research problem and an economic solution to large-scale city modeling, compared with reconstruction from LiDAR data and multi-view imagery. However, several challenges such as the partial invisibility of building footprints and facades, the serious shadow effect, and the extreme variance of building height in large-scale areas, have restricted the existing monocular image based building reconstruction studies to certain application scenes, i.e., modeling simple low-rise buildings from near-nadir images. In this study, we propose a novel 3D building reconstruction method for monocular remote sensing images, which tackles the above difficulties, thus providing an appealing solution for more complicated scenarios. We design a multi-task building reconstruction network, named MTBR-Net, to learn the geometric property of oblique images, the key components of a 3D building model and their relations via four semantic-related and three offset-related tasks. The network outputs are further integrated by a prior knowledge based 3D model optimization method to produce the the final 3D building models. Results on a public 3D reconstruction dataset and a novel released dataset demonstrate that our method improves the height estimation performance by over 40% and the segmentation F1-score by 2% - 4% compared with current state-of-the-art.

City modeling is the foundation for computational urban planning, navigation, and entertainment. In this work, we present the first generative model of city blocks named BlockPlanner, and showcase its ability to synthesize valid city blocks with varying land lots configurations. We propose a novel vectorized city block representation utilizing a ring topology and a two-tier graph to capture the global and local structures of a city block. Each land lot is abstracted into a vector representation covering both its 3D geometry and land use semantics. Such vectorized representation enables us to deploy a lightweight network to capture the underlying distribution of land lots configurations in a city block. To enforce intrinsic spatial constraints of a valid city block, a set of effective loss functions are imposed to shape rational results. We contribute a pilot city block dataset to demonstrate the effectiveness and efficiency of our representation and framework over the state-of-the-art. Notably, our BlockPlanner is also able to edit and manipulate city blocks, enabling several useful applications, e.g., topology refinement and footprint generation.

Transformers with powerful global relation modeling abilities have been introduced to fundamental computer vision tasks recently. As a typical example, the Vision Trans-former (ViT) directly applies a pure transformer architecture on image classification, by simply splitting images into tokens with a fixed length, and employing transformers to learn relations between these tokens. However, such naive tokenization could destruct object structures, assign grids to uninterested regions such as background, and introduce interference signals. To mitigate the above issues, in this paper, we propose an iterative and progressive sampling strategy to locate discriminative regions. At each iteration, embeddings of the current sampling step are fed into a transformer encoder layer, and a group of sampling off-sets is predicted to update the sampling locations for the next step. The progressive sampling is differentiable. When combined with the Vision Transformer, the obtained PS-ViT network can adaptively learn where to look. The proposed PS-ViT is both effective and efficient. When trained from scratch on ImageNet, PS-ViT performs 3.8% higher than the vanilla ViT in terms of top-1 accuracy with about 4× fewer parameters and 10× fewer FLOPs. Code is available at https://github.com/yuexy/PS-ViT.

We present MMOCR---an open-source toolbox which provides a comprehensive pipeline for text detection and recognition, as well as their downstream tasks such as named entity recognition and key information extraction. MMOCR implements 14 state-of-the-art algorithms, which is significantly more than all the existing open-source OCR projects we are aware of to date. To facilitate future research and industrial applications of text recognition-related problems, we also provide a large number of trained models and detailed benchmarks to give insights into the performance of text detection, recognition and understanding. MMOCR is publicly released at https://github.com/open-mmlab/mmocr.

Object detection has achieved substantial progress in the last decade. However, detecting novel classes with only few samples remains challenging, since deep learning under low data regime usually leads to a degraded feature space. Existing works employ a holistic fine-tuning paradigm to tackle this problem, where the model is first pre-trained on all base classes with abundant samples, and then it is used to carve the novel class feature space. Nonetheless, this paradigm is still imperfect. Durning fine-tuning, a novel class may implicitly leverage the knowledge of multiple base classes to construct its feature space, which induces a scattered feature space, hence violating the inter-class separability. To overcome these obstacles, we propose a two-step fine-tuning framework, Few-shot object detection via Association and DIscrimination (FADI), which builds up a discriminative feature space for each novel class with two integral steps. 1) In the association step, in contrast to implicitly leveraging multiple base classes, we construct a compact novel class feature space via explicitly imitating a specific base class feature space. Specifically, we associate each novel class with a base class according to their semantic similarity. After that, the feature space of a novel class can readily imitate the well-trained feature space of the associated base class. 2) In the discrimination step, to ensure the separability between the novel classes and associated base classes, we disentangle the classification branches for base and novel classes. To further enlarge the inter-class separability between all classes, a set-specialized margin loss is imposed. Extensive experiments on standard Pascal VOC and MS-COCO datasets demonstrate that FADI achieves new state-of-the-art performance, significantly improving the baseline in any shot/split by +18.7. Notably, the advantage of FADI is most announced on extremely few-shot scenarios (e.g. 1- and 3- shot).

Chamfer Distance (CD) and Earth Mover’s Distance (EMD) are two broadly adopted metrics for measuring the similarity between two point sets. However, CD is usually insensitive to mismatched local density, and EMD is usually dominated by global distribution while overlooks the fidelity of detailed structures. Besides, their unbounded value range induces a heavy influence from the outliers. These defects prevent them from providing a consistent evaluation. To tackle these problems, we propose a new similarity measure named Density-aware Chamfer Distance (DCD). It is derived from CD and benefits from several desirable properties: 1) it can detect disparity of density distributions and is thus a more intensive measure of similarity compared to CD; 2) it is stricter with detailed structures and significantly more computationally efficient than EMD; 3) the bounded value range encourages a more stable and reasonable evaluation over the whole test set. We adopt DCD to evaluate the point cloud completion task, where experimental results show that DCD pays attention to both the overall structure and local geometric details and provides a more reliable evaluation even when CD and EMD contradict each other. We can also use DCD as the training loss, which outperforms the same model trained with CD loss on all three metrics. In addition, we propose a novel point discriminator module that estimates the priority for another guided down-sampling step, and it achieves noticeable improvements under DCD together with competitive results for both CD and EMD. We hope our work could pave the way for a more comprehensive and practical point cloud similarity evaluation. Our code will be available at https://github.com/wutong16/DensityawareChamfer_Distance.

The advent of generative radiance fields has significantly promoted the development of 3D-aware image synthesis. The cumulative rendering process in radiance fields makes training these generative models much easier since gradients are distributed over the entire volume, but leads to diffused object surfaces. In the meantime, compared to radiance fields occupancy representations could inherently ensure deterministic surfaces. However, if we directly apply occupancy representations to generative models, during training they will only receive sparse gradients located on object surfaces and eventually suffer from the convergence problem. In this paper, we propose Generative Occupancy Fields (GOF), a novel model based on generative radiance fields that can learn compact object surfaces without impeding its training convergence. The key insight of GOF is a dedicated transition from the cumulative rendering in radiance fields to rendering with only the surface points as the learned surface gets more and more accurate. In this way, GOF combines the merits of two representations in a unified framework. In practice, the training-time transition of start from radiance fields and march to occupancy representations is achieved in GOF by gradually shrinking the sampling region in its rendering process from the entire volume to a minimal neighboring region around the surface. Through comprehensive experiments on multiple datasets, we demonstrate that GOF can synthesize high-quality images with 3D consistency and simultaneously learn compact and smooth object surfaces. Our code is available at https://github.com/SheldonTsui/GOF_NeurIPS2021.

Recovering three-dimensional (3D) structures such as object poses from limited two-dimensional (2D) information is an important research problem in computer vision, graphics, and robotics. The estimation of object pose from single images or multiple casual images could be ill-conditioned math problems. There is a popular family of algorithms of geometric sparse representation for 3D pose recovery (GSR-3D) that pretrains an overcomplete dictionary of 3D basis poses $B$, and then matches the detected 2D object pose $Y$ by jointly estimating the transformation $R$, projection $\Pi,$ and combination coefficients $c$, assuming $Y \approx \Pi \sum_i c_i R B_i$. In this paper, we make the first step of analyzing to which extent could we solve this ill-conditioned problem, and of understanding how the recovery error is affected by fundamental factors, e.g., dictionary size, observation noise, and running time. As these factors are implicit in objective functions, we analyze with the help of various sparse regularizers and a multistage optimizer, and prove that the recovery error $\mathcal L(l)$ decays w.r.t. the number of stages $l$ with a high probability, $Prob\left(\mathcal L(l) < \rho^{l-1} \mathcal L(0) + \delta \right) \geq 1- \epsilon$, where the constants $0< \rho <1, 0<\delta, 0<\epsilon \ll 1$ are related to the aforementioned factors. To the best of our knowledge, this is the first theoretical analysis in this line of research. Experiments are conducted to support our improvement upon previous regularization within the same framework. This will further characterize the trade-off between speed and accuracy towards real-time geometric inference in applications.

Fashion recommendation systems are highly desired by customers to find visually-collocated fashion items, such as clothes, shoes, bags, etc. While existing methods demonstrate promising results, they remain lacking in flexibility and diversity, e.g. assuming a fixed number of items or favoring safe but boring recommendations. In this paper, we propose a novel fashion collocation framework,Neural Graph Filtering, that models a flexible set of fashion items via a graph neural network. Specifically, we consider the visual embeddings of each garment as a node in the graph, and describe the inter-garment relationship as the edge between nodes. By applying symmetric operations on the edge vectors, this framework allows varying numbers of inputs/outputs and is invariant to their ordering. We further include a style classifier augmented with focal loss to enable the collocation of significantly diverse styles, which are inherently imbalanced in the training set. To facilitate a comprehensive study on diverse fashion collocation, we reorganize Amazon Fashion dataset with carefully designed evaluation protocols. We evaluate the proposed approach on three popular benchmarks, the Polyvore dataset, the Polyvore-D dataset, and our reorganized Amazon Fashion dataset. Extensive experimental results show that our approach significantly outperforms the state-of-the-art methods with over 10% improvements on the standard AUC metric. More importantly, 82.5% of the users prefer our diverse-style recommendations over other alternatives in a real-world perception study.

The rapid growth of deep learning applications in real life is accompanied by severe safety concerns. To mitigate this uneasy phenomenon, much research has been done providing reliable evaluations of the fragility level in different deep neural networks. Apart from devising adversarial attacks, quantifiers that certify safeguarded regions have also been designed in the past five years. The summarizing work in (Salman et al. 2019) unifies a family of existing verifiers under a convex relaxation framework. We draw inspiration from such work and further demonstrate the optimality of deterministic CROWN (Zhang et al. 2018) solutions in a given linear programming problem under mild constraints. Given this theoretical result, the computationally expensive linear programming based method is shown to be unnecessary. We then propose an optimization-based approach FROWN (Fastened CROWN): a general algorithm to tighten robustness certificates for neural networks. Extensive experiments on various networks trained individually verify the effectiveness of FROWN in safeguarding larger robust regions.

A typical domain adaptation approach is to adapt models trained on the annotated data in a source domain (e.g., sunny weather) for achieving high performance on the test data in a target domain (e.g., rainy weather). Whether the target contains a single homogeneous domain or multiple heterogeneous domains, existing works always assume that there exist clear distinctions between the domains, which is often not true in practice (e.g., changes in weather). We study an open compound domain adaptation (OCDA) problem, in which the target is a compound of multiple homogeneous domains without domain labels, reflecting realistic data collection from mixed and novel situations. We propose a new approach based on two technical insights into OCDA: 1) a curriculum domain adaptation strategy to bootstrap generalization across domains in a data-driven self-organizing fashion and 2) a memory module to increase the model's agility towards novel domains. Our experiments on digit classification, facial expression recognition, semantic segmentation, and reinforcement learning demonstrate the effectiveness of our approach.

It is a common paradigm in object detection frameworks to treat all samples equally and target at maximizing the performance on average. In this work, we revisit this paradigm through a careful study on how different samples contribute to the overall performance measured in terms of mAP. Our study suggests that the samples in each mini-batch are neither independent nor equally important, and therefore a better classifier on average does not necessarily result in higher mAP. Motivated by this study, we propose the notion of Prime Samples, those that play a key role in driving the detection performance. We further develop a simple yet effective sampling and learning strategy called PrIme Sample Attention (PISA) that directs the focus of the training process towards such samples. Our experiments demonstrate that it is often more effective to focus on prime samples than hard samples when training a detector. Particularly, on the MSCOCO dataset, PISA outperforms the random sampling baseline and hard mining schemes, \eg~OHEM and Focal Loss, consistently by around 2\% on both single-stage and two-stage detectors, even with a strong backbone ResNeXt-101. Code is available at: \url{https://github.com/open-mmlab/mmdetection}.

Recent advances in adversarial attacks uncover the intrinsic vulnerability of modern deep neural networks. Since then, extensive efforts have been devoted to enhancing the robustness of deep networks via specialized learning algorithms and loss functions. In this work, we take an architectural perspective and investigate the patterns of network architectures that are resilient to adversarial attacks. To obtain the large number of networks needed for this study, we adopt one-shot neural architecture search, training a large network for once and then finetuning the sub-networks sampled therefrom. The sampled architectures together with the accuracies they achieve provide a rich basis for our study. Our ''robust architecture Odyssey'' reveals several valuable observations: 1) densely connected patterns result in improved robustness; 2) under computational budget, adding convolution operations to direct connection edge is effective; 3) flow of solution procedure (FSP) matrix is a good indicator of network robustness. Based on these observations, we discover a family of robust architectures (RobNets). On various datasets, including CIFAR, SVHN, Tiny-ImageNet, and ImageNet, RobNets exhibit superior robustness performance to other widely used architectures. Notably, RobNets substantially improve the robust accuracy (~5% absolute gains) under both white-box and black-box attacks, even with fewer parameter numbers. Code is available at https://github.com/gmh14/RobNets.

If NAS methods are solutions, what is the problem? Most existing NAS methods require two-stage parameter optimization. However, performance of the same architecture in the two stages correlates poorly. In this work, we propose a new problem definition for NAS, task-specific end-to-end, based on this observation. We argue that given a computer vision task for which a NAS method is expected, this definition can reduce the vaguely-defined NAS evaluation to i) accuracy of this task and ii) the total computation consumed to finally obtain a model with satisfying accuracy. Seeing that most existing methods do not solve this problem directly, we propose DSNAS, an efficient differentiable NAS framework that simultaneously optimizes architecture and parameters with a low-biased Monte Carlo estimate. Child networks derived from DSNAS can be deployed directly without parameter retraining. Comparing with two-stage methods, DSNAS successfully discovers networks with comparable accuracy (74.4\%) on ImageNet in 420 GPU hours, reducing the total time by more than 34\%.

Scene, as the crucial unit of storytelling in movies, contains complex activities of actors and their interactions in a physical environment. Identifying the composition of scenes serves as a critical step towards semantic understanding of movies. This is very challenging - compared to the videos studied in conventional vision problems, e.g. action recognition, as scenes in movies usually contain much richer temporal structures and more complex semantic information. Towards this goal, we scale up the scene segmentation task by building a large-scale video dataset MovieScenes, which contains 21K annotated scene segments from 150 movies. We further propose a local-to-global scene segmentation framework, which integrates multi-modal information across three levels, i.e. clip, segment, and movie. This framework is able to distill complex semantics from hierarchical temporal structures over a long movie, providing top-down guidance for scene segmentation. Our experiments show that the proposed network is able to segment a movie into scenes with high accuracy, consistently outperforming previous methods. We also found that pretraining on our MovieScenes can bring significant improvements to the existing approaches.

Current methods for action recognition primarily rely on deep convolutional networks to derive feature embeddings of visual and motion features. While these methods have demonstrated remarkable performance on standard benchmarks, we are still in need of a better understanding as to how the videos, in particular their internal structures, relate to high-level semantics, which may lead to benefits in multiple aspects, e.g. interpretable predictions and even new methods that can take the recognition performances to a next level. Towards this goal, we construct TAPOS, a new dataset developed on sport videos with manual annotations of sub-actions, and conduct a study on temporal action parsing on top. Our study shows that a sport activity usually consists of multiple sub-actions and that the awareness of such temporal structures is beneficial to action recognition. We also investigate a number of temporal parsing methods, and thereon devise an improved method that is capable of mining sub-actions from training data without knowing the labels of them. On the constructed TAPOS, the proposed method is shown to reveal intra-action information, i.e. how action instances are made of sub-actions, and inter-action information, i.e. one specific sub-action may commonly appear in various actions.

On public benchmarks, current action recognition techniques have achieved great success. However, when used in real-world applications, e.g. sport analysis, which requires the capability of parsing an activity into phases and differentiating between subtly different actions, their performances remain far from being satisfactory. To take action recognition to a new level, we develop FineGym, a new dataset built on top of gymnasium videos. Compared to existing action recognition datasets, FineGym is distinguished in richness, quality, and diversity. In particular, it provides temporal annotations at both action and sub-action levels with a three-level semantic hierarchy. For example, a “balance beam” activity will be annotated as a sequence of elementary sub-actions derived from five sets: “leap-jump-hop”, “beam-turns”, “flight-salto”, “flight-handspring”, and “dismount”, where the sub-action in each set will be further annotated with finely defined class labels. This new level of granularity presents significant challenges for action recognition, e.g. how to parse the temporal structures from a coherent action, and how to distinguish between subtly different action classes. We systematically investigates different methods on this dataset and obtains a number of interesting findings. We hope this dataset could advance research towards action understanding.

Face clustering is an essential tool for exploiting the unlabeled face data, and has a wide range of applications including face annotation and retrieval. Recent works show that supervised clustering can result in noticeable performance gain. However, they usually involve heuristic steps and require numerous overlapped subgraphs, severely restricting their accuracy and efficiency. In this paper, we propose a fully learnable clustering framework without requiring a large number of overlapped subgraphs. Instead, we transform the clustering problem into two sub-problems. Specifically, two graph convolutional networks, named GCN-V and GCN-E, are designed to estimate the confidence of vertices and the connectivity of edges, respectively. With the vertex confidence and edge connectivity, we can naturally organize more relevant vertices on the affinity graph and group them into clusters. Experiments on two large-scale benchmarks show that our method significantly improves clustering accuracy and thus performance of the recognition models trained on top, yet it is an order of magnitude more efficient than existing supervised methods.

Natural scene understanding is a challenging task, particularly when encountering images of multiple objects that are partially occluded. This obstacle is given rise by varying object ordering and positioning. Existing scene understanding paradigms are able to parse only the visible parts, resulting in incomplete and unstructured scene interpretation. In this paper, we investigate the problem of scene de-occlusion, which aims to recover the underlying occlusion ordering and complete the invisible parts of occluded objects. We make the first attempt to address the problem through a novel and unified framework that recovers hidden scene structures without ordering and amodal annotations as supervisions. This is achieved via Partial Completion Network (PCNet)-mask (M) and -content (C), that learn to recover fractions of object masks and contents, respectively, in a self-supervised manner. Based on PCNet-M and PCNet-C, we devise a novel inference scheme to accomplish scene de-occlusion, via progressive ordering recovery, amodal completion and content completion. Extensive experiments on real-world scenes demonstrate the superior performance of our approach to other alternatives. Remarkably, our approach that is trained in a self-supervised manner achieves comparable results to fully-supervised methods. The proposed scene de-occlusion framework benefits many applications, including high-quality and controllable image manipulation and scene recomposition (see Fig. 1), as well as the conversion of existing modal mask annotations to amodal mask annotations.

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

In this paper, we propose an inverse reinforcement learning method for architecture search (IRLAS), which trains an agent to learn to search network structures that are topologically inspired by human-designed network. Most existing architecture search approaches totally neglect the topological characteristics of architectures, which results in complicated architecture with a high inference latency. Motivated by the fact that human-designed networks are elegant in topology with a fast inference speed, we propose a mirror stimuli function inspired by biological cognition theory to extract the abstract topological knowledge of an expert human-design network (ResNeXt). To avoid raising a too strong prior over the search space, we introduce inverse reinforcement learning to train the mirror stimuli function and exploit it as a heuristic guidance for architecture search, easily generalized to different architecture search algorithms. On CIFAR-10, the best architecture searched by our proposed IRLAS achieves 2.60% error rate. For ImageNet mobile setting, our model achieves a state-of-the-art top-1 accuracy 75.28%, while being 2~4x faster than most auto-generated architectures. A fast version of this model achieves 10% faster than MobileNetV2, while maintaining a higher accuracy.

Conventionally, deep neural networks are trained offline, relying on a large dataset prepared in advance. This paradigm is often challenged in real-world applications, e.g. online services that involve continuous streams of incoming data. Recently, incremental learning receives increasing attention, and is considered as a promising solution to the practical challenges mentioned above. However, it has been observed that incremental learning is subject to a fundamental difficulty -- catastrophic forgetting, namely adapting a model to new data often results in severe performance degradation on previous tasks or classes. Our study reveals that the imbalance between previous and new data is a crucial cause to this problem. In this work, we develop a new framework for incrementally learning a unified classifier, e.g. a classifier that treats both old and new classes uniformly. Specifically, we incorporate three components, cosine normalization, less-forget constraint, and inter-class separation, to mitigate the adverse effects of the imbalance. Experiments show that the proposed method can effectively rebalance the training process, thus obtaining superior performance compared to the existing methods. On CIFAR-100 and ImageNet, our method can reduce the classification errors by more than 6% and 13% respectively, under the incremental setting of 10 phases.

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

Region anchors are the cornerstone of modern object detection techniques. State-of-the-art detectors mostly rely on a dense anchoring scheme, where anchors are sampled uniformly over the spatial domain with a predefined set of scales and aspect ratios. In this paper, we revisit this foundational stage. Our study shows that it can be done much more effectively and efficiently. Specifically, we present an alternative scheme, named Guided Anchoring, which leverages semantic features to guide the anchoring. The proposed method jointly predicts the locations where the center of objects of interest are likely to exist as well as the scales and aspect ratios at different locations. On top of predicted anchor shapes, we mitigate the feature inconsistency with a feature adaption module. We also study the use of high-quality proposals to improve detection performance. The anchoring scheme can be seamlessly integrated into proposal methods and detectors. With Guided Anchoring, we achieve 9.1% higher recall on MS COCO with 90% fewer anchors than the RPN baseline. We also adopt Guided Anchoring in Fast R-CNN, Faster R-CNN and RetinaNet, respectively improving the detection mAP by 2.2%, 2.7% and 1.2%. Code is available at https://github.com/open-mmlab/mmdetection.

Face recognition sees remarkable progress in recent years, and its performance has reached a very high level. Taking it to a next level requires substantially larger data, which would involve prohibitive annotation cost. Hence, exploiting unlabeled data becomes an appealing alternative. Recent works have shown that clustering unlabeled faces is a promising approach, often leading to notable performance gains. Yet, how to effectively cluster, especially on a large-scale (i.e. million-level or above) dataset, remains an open question. A key challenge lies in the complex variations of cluster patterns, which make it difficult for conventional clustering methods to meet the needed accuracy. This work explores a novel approach, namely, learning to cluster instead of relying on hand-crafted criteria. Specifically, we propose a framework based on graph convolutional network, which combines a detection and a segmentation module to pinpoint face clusters. Experiments show that our method yields significantly more accurate face clusters, which, as a result, also lead to further performance gain in face recognition.

Intelligent agent naturally learns from motion. Various self-supervised algorithms have leveraged the motion cues to learn effective visual representations. The hurdle here is that motion is both ambiguous and complex, rendering previous works either suffer from degraded learning efficacy, or resort to strong assumptions on object motions. In this work, we design a new learning-from-motion paradigm to bridge these gaps. Instead of explicitly modeling the motion probabilities, we design the pretext task as a conditional motion propagation problem. Given an input image and several sparse flow guidance on it, our framework seeks to recover the full-image motion. Compared to other alternatives, our framework has several appealing properties: (1) Using sparse flow guidance during training resolves the inherent motion ambiguity, and thus easing feature learning. (2) Solving the pretext task of conditional motion propagation encourages the emergence of kinematically-sound representations that poss greater expressive power. Extensive experiments demonstrate that our framework learns structural and coherent features; and achieves state-of-the-art self-supervision performance on several downstream tasks including semantic segmentation, instance segmentation and human parsing. Furthermore, our framework is successfully extended to several useful applications such as semi-automatic pixel-level annotation.

State-of-the-art object detectors are usually trained on public datasets. They often face substantial difficulties when applied to a different domain, where the imaging condition differs significantly and the corresponding annotated data are unavailable (or expensive to acquire). A natural remedy is to adapt the model by aligning the image representations on both domains. This can be achieved, for example, by adversarial learning, and has been shown to be effective in tasks like image classification. However, we found that in object detection, the improvement obtained in this way is quite limited. An important reason is that conventional domain adaptation methods strive to align images as a whole, while object detection, by nature, focuses on local regions that may contain objects of interest. Motivated by this, we propose a novel approach to domain adaption for object detection to handle the issues in ``where to look'' and ``how to align''. Our key idea is to mine the discriminative regions, namely those that are directly pertinent to object detection, and focus on aligning them across both domains. Experiments show that the proposed method performs remarkably better than existing methods with about 4% ~ 6% improvement under various domain-shift scenarios while keeping good scalability.

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

Feature upsampling is a key operation in a number of modern convolutional network architectures, e.g. feature pyramids. Its design is critical for dense prediction tasks such as object detection and semantic/instance segmentation. In this work, we propose Content-Aware ReAssembly of FEatures (CARAFE), a universal, lightweight and highly effective operator to fulfill this goal. CARAFE has several appealing properties: (1) Large field of view. Unlike previous works (e.g. bilinear interpolation) that only exploit subpixel neighborhood, CARAFE can aggregate contextual information within a large receptive field. (2) Content-aware handling. Instead of using a fixed kernel for all samples (e.g. deconvolution), CARAFE enables instance-specific content-aware handling, which generates adaptive kernels on-the-fly. (3) Lightweight and fast to compute. CARAFE introduces little computational overhead and can be readily integrated into modern network architectures. We conduct comprehensive evaluations on standard benchmarks in object detection, instance/semantic segmentation and inpainting. CARAFE shows consistent and substantial gains across all the tasks (1.2% AP, 1.3% AP, 1.8% mIoU, 1.1dB respectively) with negligible computational overhead. It has great potential to serve as a strong building block for future research. Code and models are available at https://github.com/open-mmlab/mmdetection.

Inspired by the remarkable advances in video analytics, research teams are stepping towards a greater ambition - movie understanding. However, compared to those activity videos in conventional datasets, movies are significantly different. Generally, movies are much longer and consist of much richer temporal structures. More importantly, the interactions among characters play a central role in expressing the underlying story. To facilitate the efforts along this direction, we construct a dataset called Movie Synopses Associations (MSA) over 327 movies, which provides a synopsis for each movie, together with annotated associations between synopsis paragraphs and movie segments. On top of this dataset, we develop a framework to perform matching between movie segments and synopsis paragraphs. This framework integrates different aspects of a movie, including event dynamics and character interactions, and allows them to be matched with parsed paragraphs, based on a graph-based formulation. Our study shows that the proposed framework remarkably improves the matching accuracy over conventional feature-based methods. It also reveals the importance of narrative structures and character interactions in movie understanding. Dataset and code are available at: https://ycxioooong.github.io/projects/moviesyn.

Sounds provide rich semantics, complementary to visual data, for many tasks. However, in practice, sounds from multiple sources are often mixed together. In this paper we propose a novel framework, referred to as MinusPlus Network (MP-Net), for the task of visual sound separation. MP-Net separates sounds recursively in the order of average energy, removing the separated sound from the mixture at the end of each prediction, until the mixture becomes empty or contains only noise. In this way, MP-Net could be applied to sound mixtures with arbitrary numbers and types of sounds. Moreover, while MP-Net keeps removing sounds with large energy from the mixture, sounds with small energy could emerge and become clearer, so that the separation is more accurate. Compared to previous methods, MP-Net obtains state-of-the-art results on two large scale datasets, across mixtures with different types and numbers of sounds.

In this work, we aim to generate long actions represented as sequences of skeletons. The generated sequences must demonstrate continuous, meaningful human actions, while maintaining coherence among body parts. Instead of generating skeletons sequentially following an autoregressive model, we propose a framework that generates the entire sequence altogether by transforming from a sequence of latent vectors sampled from a Gaussian process (GP). This framework, named Convolutional Sequence Generation Network (CSGN), jointly models structures in temporal and spatial dimensions. It captures the temporal structure at multiple scales through the GP prior and the temporal convolutions; and establishes the spatial connection between the latent vectors and the skeleton graphs via a novel graph refining scheme. It is noteworthy that CSGN allows bidirectional transforms between the latent and the observed spaces, thus enabling semantic manipulation of the action sequences in various forms. We conducted empirical studies on multiple datasets, including a set of high-quality dancing sequences collected by us. The results show that our framework can produce long action sequences that are coherent across time steps and among body parts.

The vulnerability to adversarial attacks has been a critical issue for deep neural networks. Addressing this issue requires a reliable way to evaluate the robustness of a network. Recently, several methods have been developed to compute robustness quantification for neural networks, namely, certified lower bounds of the minimum adversarial perturbation. Such methods, however, were devised for feed-forward networks, e.g. multi-layer perceptron or convolutional networks. It remains an open problem to quantify robustness for recurrent networks, especially LSTM and GRU. For such networks, there exist additional challenges in computing the robustness quantification, such as handling the inputs at multiple steps and the interaction between gates and states. In this work, we propose POPQORN (Propagated-output Quantified Robustness for RNNs), a general algorithm to quantify robustness of RNNs, including vanilla RNNs, LSTMs, and GRUs. We demonstrate its effectiveness on different network architectures and show that the robustness quantification on individual steps can lead to new insights.

Solving goal-oriented tasks is an important but challenging problem in reinforcement learning (RL). For such tasks, the rewards are often sparse, making it difficult to learn a policy effectively. To tackle this difficulty, we propose a new approach called Policy Continuation with Hindsight Inverse Dynamics (PCHID). This approach learns from Hindsight Inverse Dynamics based on Hindsight Experience Replay. Enabling the learning process in a self-imitated manner and thus can be trained with supervised learning. This work also extends it to multi-step settings with Policy Continuation. The proposed method is general, which can work in isolation or be combined with other on-policy and off-policy algorithms. On two multi-goal tasks GridWorld and FetchReach, PCHID significantly improves the sample efficiency as well as the final performance.

We present a general and flexible video-level framework for learning action models in videos. This method, called temporal segment network (TSN), aims to model long-range temporal structure with a new segment-based sampling and aggregation scheme. This unique design enables the TSN framework to efficiently learn action models by using the whole video. The learned models could be easily deployed for action recognition in both trimmed and untrimmed videos with simple average pooling and multi-scale temporal window integration, respectively. We also study a series of good practices for the implementation of the TSN framework given limited training samples. Our approach obtains the state-the-of-art performance on five challenging action recognition benchmarks: HMDB51 (71.0 percent), UCF101 (94.9 percent), THUMOS14 (80.1 percent), ActivityNet v1.2 (89.6 percent), and Kinetics400 (75.7 percent). In addition, using the proposed RGB difference as a simple motion representation, our method can still achieve competitive accuracy on UCF101 (91.0 percent) while running at 340 FPS. Furthermore, based on the proposed TSN framework, we won the video classification track at the ActivityNet challenge 2016 among 24 teams.

Environment perception is an important task with great practical value and bird view is an essential part for creating panoramas of surrounding environment. Due to the large gap and severe deformation between the frontal view and bird view, generating a bird view image from a single frontal view is challenging. To tackle this problem, we propose the BridgeGAN, i.e., a novel generative model for bird view synthesis. First, an intermediate view, i.e., homography view, is introduced to bridge the large gap. Next, conditioned on the three views (frontal view, homography view and bird view) in our task, a multi-GAN based model is proposed to learn the challenging cross-view translation. Furthermore, to guarantee one-to-one cross-view correspondences and consistent cross-view feature representations, two consistency constraints are designed for our task. Extensive experiments conducted on a synthetic dataset have demonstrated that the images generated by our model are much better than those generated by existing methods, with more consistent global appearance and sharper details. Ablation studies and discussions show its reliability and robustness in some challenging cases.

Dynamics of human body skeletons convey significant information for human action recognition. Conventional approaches for modeling skeletons usually rely on hand-crafted parts or traversal rules, thus resulting in limited expressive power and difficulties of generalization. In this work, we propose a novel model of dynamic skeletons called Spatial-Temporal Graph Convolutional Networks (ST-GCN), which moves beyond the limitations of previous methods by automatically learning both the spatial and temporal patterns from data. This formulation not only leads to greater expressive power but also stronger generalization capability. On two large datasets, Kinetics and NTU-RGBD, it achieves substantial improvements over mainstream methods.

Massive classification, a classification task defined over a vast number of classes (hundreds of thousands or even millions), has become an essential part of many real-world systems, such as face recognition. Existing methods, including the deep networks that achieved remarkable success in recent years, were mostly devised for problems with a moderate number of classes. They would meet with substantial difficulties, e.g., excessive memory demand and computational cost, when applied to massive problems. We present a new method to tackle this problem. This method can efficiently and accurately identify a small number of "active classes" for each mini-batch, based on a set of dynamic class hierarchies constructed on the fly. We also develop an adaptive allocation scheme thereon, which leads to a better tradeoff between performance and cost. On several large-scale benchmarks, our method significantly reduces the training cost and memory demand, while maintaining competitive performance.

High-performance object detection relies on expensive convolutional networks to compute features, often leading to significant challenges in applications, e.g. those that require detecting objects from video streams in real time. The key to this problem is to trade accuracy for efficiency in an effective way, i.e. reducing the computing cost while maintaining competitive performance. To seek a good balance, previous efforts usually focus on optimizing the model architectures. This paper explores an alternative approach, that is, to reallocate the computation over a scale-time space. The basic idea is to perform expensive detection sparsely and propagate the results across both scales and time with substantially cheaper networks, by exploiting the strong correlations among them. Specifically, we present a unified framework that integrates detection, temporal propagation, and across-scale refinement on a Scale-Time Lattice. On this framework, one can explore various strategies to balance performance and cost. Taking advantage of this flexibility, we further develop an adaptive scheme with the detector invoked on demand and thus obtain improved tradeoff. On ImageNet VID dataset, the proposed method can achieve a competitive mAP 79.6% at 20 fps, or 79.0% at 62 fps as a performance/speed tradeoff.1

Despite the great success of face recognition techniques, recognizing persons under unconstrained settings remains challenging. Issues like profile views, unfavorable lighting, and occlusions can cause substantial difficulties. Previous works have attempted to tackle this problem by exploiting the context, e.g. clothes and social relations. While showing promising improvement, they are usually limited in two important aspects, relying on simple heuristics to combine different cues and separating the construction of context from people identities. In this work, we aim to move beyond such limitations and propose a new framework to leverage context for person recognition. In particular, we propose a Region Attention Network, which is learned to adaptively combine visual cues with instance-dependent weights. We also develop a unified formulation, where the social contexts are learned along with the reasoning of people identities. These models substantially improve the robustness when working with the complex contextual relations in unconstrained environments. On two large datasets, PIPA [27] and Cast In Movies (CIM), a new dataset proposed in this work, our method consistently achieves state-of-the-art performance under multiple evaluation policies.

Recent years have seen remarkable progress in semantic segmentation. Yet, it remains a challenging task to apply segmentation techniques to video-based applications. Specifically, the high throughput of video streams, the sheer cost of running fully convolutional networks, together with the low-latency requirements in many real-world applications, e.g. autonomous driving, present a significant challenge to the design of the video segmentation framework. To tackle this combined challenge, we develop a framework for video semantic segmentation, which incorporates two novel components: (1) a feature propagation module that adaptively fuses features over time via spatially variant convolution, thus reducing the cost of per-frame computation: and (2) an adaptive scheduler that dynamically allocate computation based on accuracy prediction. Both components work together to ensure low latency while maintaining high segmentation quality. On both Cityscapes and CamVid, the proposed framework obtained competitive performance compared to the state of the art, while substantially reducing the latency, from 360 ms to 119 ms.

In this paper, we propose a simple yet effective method to learn globally optimized detector for object detection, which is a simple modification to the standard cross-entropy gradient inspired by the REINFORCE algorithm. In our approach, the cross-entropy gradient is adaptively adjusted according to overall mean Average Precision (mAP) of the current state for each detection candidate, which leads to more effective gradient and global optimization of detection results, and brings no computational overhead. Benefiting from more precise gradients produced by the global optimization method, our framework significantly improves state-of-the-art object detectors. Furthermore, since our method is based on scores and bounding boxes without modification on the architecture of object detector, it can be easily applied to off-the-shelf modern object detection frameworks.

Neural net classifiers trained on data with annotated class labels can also capture apparent visual similarity among categories without being directed to do so. We study whether this observation can be extended beyond the conventional domain of supervised learning: Can we learn a good feature representation that captures apparent similarity among instances, instead of classes, by merely asking the feature to be discriminative of individual instances? We formulate this intuition as a non-parametric classification problem at the instance-level, and use noise-contrastive estimation to tackle the computational challenges imposed by the large number of instance classes. Our experimental results demonstrate that, under unsupervised learning settings, our method surpasses the state-of-the-art on ImageNet classification by a large margin. Our method is also remarkable for consistently improving test performance with more training data and better network architectures. By fine-tuning the learned feature, we further obtain competitive results for semi-supervised learning and object detection tasks. Our non-parametric model is highly compact: With 128 features per image, our method requires only 600MB storage for a million images, enabling fast nearest neighbour retrieval at the run time.

Despite the remarkable progress in action recognition over the past several years, existing methods remain limited in efficiency and effectiveness. The methods treating appearance and motion as separate streams are usually subject to the cost of optical flow computation, while those relying on 3D convolution on the original video frames often yield inferior performance in practice. In this paper, we propose a new ConvNet architecture for video representation learning, which can derive disentangled components of dynamics purely from raw video frames, without the need of optical flow estimation. Particularly, the learned representation comprises three components for representing static appearance, apparent motion, and appearance changes. We introduce 3D pooling, cost volume processing, and warped feature differences, respectively for extracting the three components above. These modules are incorporated as three branches in our unified network, which share the underlying features and are learned jointly in an end-to-end manner. On two large datasets, UCF101 [22] and Kinetics [16], our method obtained competitive performances with high efficiency, using only the RGB frame sequence as input.

Mainstream captioning models often follow a sequential structure to generate cap- tions, leading to issues such as introduction of irrelevant semantics, lack of diversity in the generated captions, and inadequate generalization performance. In this paper, we present an alternative paradigm for image captioning, which factorizes the captioning procedure into two stages: (1) extracting an explicit semantic representation from the given image; and (2) constructing the caption based on a recursive compositional procedure in a bottom-up manner. Compared to conventional ones, our paradigm better preserves the semantic content through an explicit factorization of semantics and syntax. By using the compositional generation procedure, caption construction follows a recursive structure, which naturally fits the properties of human language. Moreover, the proposed compositional procedure requires less data to train, generalizes better, and yields more diverse captions.

How to leverage the temporal dimension is a key question in video analysis. Recent works suggest an efficient approach to video feature learning, i.e., factorizing 3D convolutions into separate components respectively for spatial and temporal convolutions. The temporal convolution, however, comes with an implicit assumption – the feature maps across time steps are well aligned so that the features at the same locations can be aggregated. This assumption may be overly strong in practical applications, especially in action recognition where the motion serves as a crucial cue. In this work, we propose a new CNN architecture TrajectoryNet, which incorporates trajectory convolution, a new operation for integrating features along the temporal dimension, to replace the existing temporal convolution. This operation explicitly takes into account the changes in contents caused by deformation or motion, allowing the visual features to be aggregated along the the motion paths, trajectories. On two large-scale action recognition datasets, namely, Something-Something and Kinetics, the proposed network architecture achieves notable improvement over strong baselines.

Despite the remarkable progress in recent years, detecting objects in a new context remains a challenging task. Detectors learned from a public dataset can only work with a fixed list of categories, while training from scratch usually requires a large amount of training data with detailed annotations. This work aims to explore a novel approach – learning object detectors from documentary films in a weakly supervised manner. This is inspired by the observation that documentaries often provide dedicated exposition of certain object categories, where visual presentations are aligned with subtitles. We believe that object detectors can be learned from such a rich source of information. Towards this goal, we develop a joint probabilistic framework, where individual pieces of information, including video frames and subtitles, are brought together via both visual and linguistic links. On top of this formulation, we further derive a weakly supervised learning algorithm, where object model learning and training set mining are unified in an optimization procedure. Experimental results on a real world dataset demonstrate that this is an effective approach to learning new object detectors.

Relationships among objects play a crucial role in image understanding. Despite the great success of deep learning techniques in recognizing individual objects, reasoning about the relationships among objects remains a challenging task. Previous methods often treat this as a classification problem, considering each type of relationship (e.g. ride) or each distinct visual phrase (e.g. person-ride-horse) as a category. Such approaches are faced with significant difficulties caused by the high diversity of visual appearance for each kind of relationships or the large number of distinct visual phrases. We propose an integrated framework to tackle this problem. At the heart of this framework is the Deep Relational Network, a novel formulation designed specifically for exploiting the statistical dependencies between objects and their relationships. On two large data sets, the proposed method achieves substantial improvement over state-of-the-art.

Current action recognition methods heavily rely on trimmed videos for model training. However, it is expensive and time-consuming to acquire a large-scale trimmed video dataset. This paper presents a new weakly supervised architecture, called UntrimmedNet, which is able to directly learn action recognition models from untrimmed videos without the requirement of temporal annotations of action instances. Our UntrimmedNet couples two important components, the classification module and the selection module, to learn the action models and reason about the temporal duration of action instances, respectively. These two components are implemented with feed-forward networks, and UntrimmedNet is therefore an end-to-end trainable architecture. We exploit the learned models for action recognition (WSR) and detection (WSD) on the untrimmed video datasets of THUMOS14 and ActivityNet. Although our UntrimmedNet only employs weak supervision, our method achieves performance superior or comparable to that of those strongly supervised approaches on these two datasets.

A number of studies have shown that increasing the depth or width of convolutional networks is a rewarding approach to improve the performance of image recognition. In our study, however, we observed difficulties along both directions. On one hand, the pursuit for very deep networks is met with a diminishing return and increased training difficulty, on the other hand, widening a network would result in a quadratic growth in both computational cost and memory demand. These difficulties motivate us to explore structural diversity in designing deep networks, a new dimension beyond just depth and width. Specifically, we present a new family of modules, namely the PolyInception, which can be flexibly inserted in isolation or in a composition as replacements of different parts of a network. Choosing PolyInception modules with the guidance of architectural efficiency can improve the expressive power while preserving comparable computational cost. The Very Deep PolyNet, designed following this direction, demonstrates substantial improvements over the state-of-the-art on the ILSVRC 2012 benchmark. Compared to Inception-ResNet-v2, it reduces the top-5 validation error on single crops from 4.9% to 4.25%, and that on multi-crops from 3.7% to 3.45%.

Despite the substantial progress in recent years, the image captioning techniques are still far from being perfect. Sentences produced by existing methods, e.g. those based on RNNs, are often overly rigid and lacking in variability. This issue is related to a learning principle widely used in practice, that is, to maximize the likelihood of training samples. This principle encourages high resemblance to the “ground-truth” captions, while suppressing other reasonable descriptions. Conventional evaluation metrics, e.g. BLEU and METEOR, also favor such restrictive methods. In this paper, we explore an alternative approach, with the aim to improve the naturalness and diversity - two essential properties of human expression. Specifically, we propose a new framework based on Conditional Generative Adversarial Networks (CGAN), which jointly learns a generator to produce descriptions conditioned on images and an evaluator to assess how well a description fits the visual content. It is noteworthy that training a sequence generator is nontrivial. We overcome the difficulty by Policy Gradient, a strategy stemming from Reinforcement Learning, which allows the generator to receive early feedback along the way. We tested our method on two large datasets, where it performed competitively against real people in our user study and outperformed other methods on various tasks.

Detecting actions in untrimmed videos is an important yet challenging task. In this paper, we present the structured segment network (SSN), a novel framework which models the temporal structure of each action instance via a structured temporal pyramid. On top of the pyramid, we further introduce a decomposed discriminative model comprising two classifiers, respectively for classifying actions and determining completeness. This allows the framework to effectively distinguish positive proposals from background or incomplete ones, thus leading to both accurate recognition and localization. These components are integrated into a unified network that can be efficiently trained in an end-to-end fashion. Additionally, a simple yet effective temporal action proposal scheme, dubbed temporal actionness grouping (TAG) is devised to generate high quality action proposals. On two challenging benchmarks, THUMOS14 and ActivityNet, our method remarkably outperforms previous state-of-the-art methods, demonstrating superior accuracy and strong adaptivity in handling actions with various temporal structures.

We present a novel and effective approach for generating new clothing on a wearer through generative adversarial learning. Given an input image of a person and a sentence describing a different outfit, our model “redresses” the person as desired, while at the same time keeping the wearer and her/his pose unchanged. Generating new outfits with precise regions conforming to a language description while retaining wearer's body structure is a new challenging task. Existing generative adversarial networks are not ideal in ensuring global coherence of structure given both the input photograph and language description as conditions. We address this challenge by decomposing the complex generative process into two conditional stages. In the first stage, we generate a plausible semantic segmentation map that obeys the wearer's pose as a latent spatial arrangement. An effective spatial constraint is formulated to guide the generation of this semantic segmentation map. In the second stage, a generative model with a newly proposed compositional mapping layer is used to render the final image with precise regions and textures conditioned on this map. We extended the DeepFashion dataset [8] by collecting sentence descriptions for 79K images. We demonstrate the effectiveness of our approach through both quantitative and qualitative evaluations. A user study is also conducted.

Specialized classifiers, namely those dedicated to a subset of classes, are often adopted in real-world recognition systems. However, integrating such classifiers is nontrivial. Existing methods, e.g. weighted average, usually implicitly assume that all constituents of an ensemble cover the same set of classes. Such methods can produce misleading predictions when used to combine specialized classifiers. This work explores a novel approach. Instead of combining predictions from individual classifiers directly, it first decomposes the predictions into sets of pairwise preferences, treating them as transition channels between classes, and thereon constructs a continuous-time Markov chain, and use the equilibrium distribution of this chain as the final prediction. This way allows us to form a coherent picture over all specialized predictions. On large public datasets, the proposed method obtains considerable improvement compared to mainstream ensemble methods, especially when the classifier coverage is highly unbalanced.

We consider the estimation of Dirichlet Process Mixture Models (DPMMs) in distributed environments, where data are distributed across multiple computing nodes. A key advantage of Bayesian nonparametric models such as DPMMs is that they allow new components to be introduced on the fly as needed. This, however, posts an important challenge to distributed estimation -- how to handle new components efficiently and consistently. To tackle this problem, we propose a new estimation method, which allows new components to be created locally in individual computing nodes. Components corresponding to the same cluster will be identified and merged via a probabilistic consolidation scheme. In this way, we can maintain the consistency of estimation with very low communication cost. Experiments on large real-world data sets show that the proposed method can achieve high scalability in distributed and asynchronous environments without compromising the mixing performance.

Image captioning, a popular topic in computer vision, has achieved substantial progress in recent years. However, the distinctiveness of natural descriptions is often overlooked in previous work. It is closely related to the quality of captions, as distinctive captions are more likely to describe images with their unique aspects. In this work, we propose a new learning method, Contrastive Learning (CL), for image captioning. Specifically, via two constraints formulated on top of a reference model, the proposed method can encourage distinctiveness, while maintaining the overall quality of the generated captions. We tested our method on two challenging datasets, where it improves the baseline model by significant margins. We also showed in our studies that the proposed method is generic and can be used for models with various structures.

The rapid growth of web images presents new challenges as well as opportunities to the task of image understanding. Conventional approaches rely heavily on fine-grained annotations, such as bounding boxes and semantic segmentations, which are not available for web-scale images. In general, images over the Internet are accompanied with descriptive texts, which are relevant to their contents. To bridge the gap between textual and visual analysis for image understanding, this paper presents an algorithm to learn the relations between scenes, objects, and texts with the help of image-level annotations. In particular, the relation between the texts and objects is modeled as the matching probability between the nouns and the object classes, which can be solved via a constrained bipartite matching problem. On the other hand, the relations between the scenes and objects/texts are modeled as the conditional distributions of their co-occurrence. Built upon the learned cross-domain relations, an integrated model brings together scenes, objects, and texts for joint image understanding, including scene classification, object classification and localization, and the prediction of object cardinalities. The proposed cross-domain learning algorithm and the integrated model elevate the performance of image understanding for web images in the context of textual descriptions. Experimental results show that the proposed algorithm significantly outperforms conventional methods in various computer vision tasks.

This paper proposes a novel framework for generating lingual descriptions of indoor scenes. Whereas substantial efforts have been made to tackle this problem, previous ap-proaches focusing primarily on generating a single sentence for each image, which is not sufficient for describing complex scenes. We attempt to go beyond this, by generating co-herent descriptions with multiple sentences. Our approach is distinguished from conven-tional ones in several aspects: (1) a 3D visual parsing system that jointly infers objects, attributes, and relations; (2) a generative grammar learned automatically from training text; and (3) a text generation algorithm that takes into account coherence among sen-tences. Experiments on the NYU-v2 dataset show that our framework is able to generate natural multi-sentence descriptions, outperforming those produced by a baseline. 1

A considerable portion of web images capture events that occur in our personal lives or social activities. In this paper, we aim to develop an effective method for recognizing events from such images. Despite the sheer amount of study on event recognition, most existing methods rely on videos and are not directly applicable to this task. Generally, events are complex phenomena that involve interactions among people and objects, and therefore analysis of event photos requires techniques that can go beyond recognizing individual objects and carry out joint reasoning based on evidences of multiple aspects. Inspired by the recent success of deep learning, we formulate a multi-layer framework to tackle this problem, which takes into account both visual appearance and the interactions among humans and objects, and combines them via semantic fusion. An important issue arising here is that humans and objects discovered by detectors are in the form of bounding boxes, and there is no straightforward way to represent their interactions and incorporate them with a deep network. We address this using a novel strategy that projects the detected instances onto multi-scale spatial maps. On a large dataset with 60, 000 images, the proposed method achieved substantial improvement over the state-of-the-art, raising the accuracy of event recognition by over 10%.

In this paper we exploit natural sentential descriptions of RGB-D scenes in order to improve 3D semantic parsing. Importantly, in doing so, we reason about which particular object each noun/pronoun is referring to in the image. This allows us to utilize visual information in order to disambiguate the so-called coreference resolution problem that arises in text. Towards this goal, we propose a structure prediction model that exploits potentials computed from text and RGB-D imagery to reason about the class of the 3D objects, the scene type, as well as to align the nouns/pronouns with the referred visual objects. We demonstrate the effectiveness of our approach on the challenging NYU-RGBD v2 dataset, which we enrich with natural lingual descriptions. We show that our approach significantly improves 3D detection and scene classification accuracy, and is able to reliably estimate the text-to-image alignment. Furthermore, by using textual and visual information, we are also able to successfully deal with coreference in text, improving upon the state-of-the-art Stanford coreference system [15].

In this paper, we tackle the problem of retrieving videos using complex natural language queries. Towards this goal, we first parse the sentential descriptions into a semantic graph, which is then matched to visual concepts using a generalized bipartite matching algorithm. Our approach exploits object appearance, motion and spatial relations, and learns the importance of each term using structure prediction. We demonstrate the effectiveness of our approach on a new dataset designed for semantic search in the context of autonomous driving, which exhibits complex and highly dynamic scenes with many objects. We show that our approach is able to locate a major portion of the objects described in the query with high accuracy, and improve the relevance in video retrieval.

Images are often used to convey many different concepts or illustrate many different stories. We propose an algorithm to mine multiple diverse, relevant, and interesting text snippets for images on the web. Our algorithm scales to all images on the web. For each image, all webpages that contain it are considered. The top-K text snippet selection problem is posed as combinatorial subset selection with the goal of choosing an optimal set of snippets that maximizes a combination of relevancy, interestingness, and diversity. The relevancy and interestingness are scored by machine learned models. Our algorithm is run at scale on the entire image index of a major search engine resulting in the construction of a database of images with their corresponding text snippets. We validate the quality of the database through a large-scale comparative study. We showcase the utility of the database through two web-scale applications: (a) augmentation of images on the web as webpages are browsed and (b)~an image browsing experience (similar in spirit to web browsing) that is enabled by interconnecting semantically related images (which may not be visually related) through shared concepts in their corresponding text snippets.

Age invariant face recognition has received increasing attention due to its great potential in real world applications. In spite of the great progress in face recognition techniques, reliably recognizing faces across ages remains a difficult task. The facial appearance of a person changes substantially over time, resulting in significant intra-class variations. Hence, the key to tackle this problem is to separate the variation caused by aging from the person-specific features that are stable. Specifically, we propose a new method, called Hidden Factor Analysis (HFA). This method captures the intuition above through a probabilistic model with two latent factors: an identity factor that is age-invariant and an age factor affected by the aging process. Then, the observed appearance can be modeled as a combination of the components generated based on these factors. We also develop a learning algorithm that jointly estimates the latent factors and the model parameters using an EM procedure. Extensive experiments on two well-known public domain face aging datasets: MORPH (the largest public face aging database) and FGNET, clearly show that the proposed method achieves notable improvement over state-of-the-art algorithms.

In this paper, we tackle the problem of indoor scene understanding using RGBD data. Towards this goal, we propose a holistic approach that exploits 2D segmentation, 3D geometry, as well as contextual relations between scenes and objects. Specifically, we extend the CPMC [3] framework to 3D in order to generate candidate cuboids, and develop a conditional random field to integrate information from different sources to classify the cuboids. With this formulation, scene classification and 3D object recognition are coupled and can be jointly solved through probabilistic inference. We test the effectiveness of our approach on the challenging NYU v2 dataset. The experimental results demonstrate that through effective evidence integration and holistic reasoning, our approach achieves substantial improvement over the state-of-the-art.

In this paper, we develop a generative model to describe the layouts of outdoor scenes - the spatial configuration of regions. Specifically, the layout of an image is represented as a composite of regions, each associated with a semantic topic. At the heart of this model is a novel stochastic process called Spatial Topic Process, which generates a spatial map of topics from a set of coupled Gaussian processes, thus allowing the distributions of topics to vary continuously across the image plane. A key aspect that distinguishes this model from previous ones consists in its capability of capturing dependencies across both locations and topics while allowing substantial variations in the layouts. We demonstrate the practical utility of the proposed model by testing it on scene classification, semantic segmentation, and layout hallucination.

Reliance on computationally expensive algorithms for inference has been limiting the use of Bayesian nonparametric models in large scale applications. To tackle this problem, we propose a Bayesian learning algorithm for DP mixture models. Instead of following the conventional paradigm -- random initialization plus iterative update, we take an progressive approach. Starting with a given prior, our method recursively transforms it into an approximate posterior through sequential variational approximation. In this process, new components will be incorporated on the fly when needed. The algorithm can reliably estimate a DP mixture model in one pass, making it particularly suited for applications with massive data. Experiments on both synthetic data and real datasets demonstrate remarkable improvement on efficiency -- orders of magnitude speed-up compared to the state-of-the-art.

We present a new generative image model, integrating techniques arising from two different domains: manifold modeling and Markov random fields. First, we develop a probabilistic model with a mixture of hyperplanes to approximate the manifold of orientable image patches, and demonstrate that it is more effective than the field of experts in expressing local texture patterns. Next, we develop a construction that yields an MRF for coherent image generation, given a configuration of local patch models, and thereby establish a prior distribution over an MRF space. Taking advantage of the model structure, we derive a variational inference algorithm, and apply it to low-level vision. In contrast to previous methods that rely on a single MRF, the method infers an approximate posterior distribution of MRFs, and recovers the underlying images by combining the predictions in a Bayesian fashion. Experiments quantitatively demonstrate superior performance as compared to state-of-the-art methods on image denoising and inpainting.

Markov random fields play a central role in solving a variety of low level vision problems, including denoising, in-painting, segmentation, and motion estimation. Much previous work was based on MRFs with hand-crafted networks, yet the underlying graphical structure is rarely explored. In this paper, we show that if appropriately estimated, the MRF's graphical structure, which captures significant information about appearance and motion, can provide crucial guidance to low level vision tasks. Motivated by this observation, we propose a principled framework to solve low level vision tasks via an exponential family of MRFs with variable structures, which we call Switchable MRFs. The approach explicitly seeks a structure that optimally adapts to the image or video along the pursuit of task-specific goals. Through theoretical analysis and experimental study, we demonstrate that the proposed method addresses a number of drawbacks suffered by previous methods, including failure to capture heavy-tail statistics, computational difficulties, and lack of generality.

Mixture distributions are often used to model complex data. In this paper, we develop a new method that jointly estimates mixture models over multiple data sets by exploiting the statistical dependencies between them. Specifically, we introduce a set of latent Dirichlet processes as sources of component models (atoms), and for each data set, we construct a nonparametric mixture model by combining sub-sampled versions of the latent DPs. Each mixture model may acquire atoms from different latent DPs, while each atom may be shared by multiple mixtures. This multi-to-multi association distinguishes the proposed method from prior constructions that rely on tree or chain structures, allowing mixture models to be coupled more flexibly. In addition, we derive a sampling algorithm that jointly infers the model parameters and present experiments on both document analysis and image modeling.

We propose a principled framework to model persistent motion in dynamic scenes. In contrast to previous efforts on object tracking and optical flow estimation that focus on local motion, we primarily aim at inferring a global model of persistent and collective dynamics. With this in mind, we first introduce the concept of geometric flow that describes motion simultaneously over space and time, and derive a vector space representation based on Lie algebra. We then extend it to model complex motion by combining multiple flows in a geometrically consistent manner. Taking advantage of the linear nature of this representation, we formulate a stochastic flow model, and incorporate a Gaussian process to capture the spatial coherence more effectively. This model leads to an efficient and robust algorithm that can integrate both point pairs and frame differences in motion estimation. We conducted experiments on different types of videos. The results clearly demonstrate that the proposed approach is effective in modeling persistent motion.

We present a novel method for constructing dependent Dirichlet processes. The approach exploits the intrinsic relationship between Dirichlet and Poisson pro- cesses in order to create a Markov chain of Dirichlet processes suitable for use as a prior over evolving mixture models. The method allows for the creation, re- moval, and location variation of component models over time while maintaining the property that the random measures are marginally DP distributed. Addition- ally, we derive a Gibbs sampling algorithm for model inference and test it on both synthetic and real data. Empirical results demonstrate that the approach is effec- tive in estimating dynamically varying mixture models.

We present a novel method for modeling dynamic visual phenomena, which consists of two key aspects. First, the integral motion of constituent elements in a dynamic scene is captured by a common underlying geometric transform process. Second, a Lie algebraic representation of the transform process is introduced, which maps the transformation group to a vector space, and thus overcomes the difficulties due to the group structure. Consequently, the statistical learning techniques based on vector spaces can be readily applied. Moreover, we discuss the intrinsic connections between the Lie algebra and the Linear dynamical processes, showing that our model induces spatially varying fields that can be estimated from local motions without continuous tracking. Following this, we further develop a statistical framework to robustly learn the flow models from noisy and partially corrupted observations. The proposed methodology is demonstrated on real world phenomenon, inferring common motion patterns from surveillance videos of crowded scenes and satellite data of weather evolution.

In this paper, we develop a new framework for face recognition based on nonparametric discriminant analysis (NDA) and multi-classifier integration. Traditional LDA-based methods suffer a fundamental limitation originating from the parametric nature of scatter matrices, which are based on the Gaussian distribution assumption. The performance of these methods notably degrades when the actual distribution is Non-Gaussian. To address this problem, we propose a new formulation of scatter matrices to extend the two-class nonparametric discriminant analysis to multi-class cases. Then, we develop two more improved multi-class NDA-based algorithms (NSA and NFA) with each one having two complementary methods based on the principal space and the null space of the intra-class scatter matrix respectively. Comparing to the NSA, the NFA is more effective in the utilization of the classification boundary information. In order to exploit the complementary nature of the two kinds of NFA (PNFA and NNFA), we finally develop a dual NFA-based multi-classifier fusion framework by employing the over complete Gabor representation to boost the recognition performance. We show the improvements of the developed new algorithms over the traditional subspace methods through comparative experiments on two challenging face databases, Purdue AR database and XM2VTS database.

Outdoor face recognition is among the most challenging problems for face recognition. In this paper, we develop a discriminant mutual subspace learning algorithm for indoor and outdoor face recognition. Unlike traditional algorithms using one subspace to model both indoor and outdoor face images, our algorithm simultaneously learn two related subspaces for indoor and outdoor images respectively thus can better model both. To further improve the recognition performance we develop a DMSL-based multi-classifier fusion framework on Gabor images using a new fusion method called adaptive informative fusion scheme. Experimental results clearly show that this framework can greatly enhance the recognition performance.

This paper presents a framework to automatically detect and recover the occluded facial region. We first derive a Bayesian formulation unifying the occlusion detection and recovery stages. Then a quality assessment model is developed to drive both the detection and recovery processes, which captures the face priors in both global correlation and local patterns. Based on this formulation, we further propose GraphCut-based detection and confidence-oriented sampling to attain optimal detection and recovery respectively. Compared to traditional works in image repairing, our approach is distinct in three aspects: (1) it frees the user from marking the occlusion area by incorporating an automatic occlusion detector; (2) it learns a face quality model as a criterion to guide the whole procedure; (3) it couples the detection and occlusion stages to simultaneously achieve two goals: accurate occlusion detection and high quality recovery. The comparative experiments show that our method can recover the occluded faces with both the global coherence and local details well preserved.

Human faces manifest distinct structures and characteristics when observed in different scales. Traditional face recognition techniques mainly rely on low-resolution face images, leading to the lost of significant information contained in the microscopic traits. In this paper, we introduce a multilayer framework for high resolution face recognition exploiting features in multiple scales. Each face image is factorized into four layers: global appearance, facial organs, skins, and irregular details. We employ Multilevel PCA followed by Regularized LDA to model global appearance and facial organs. However, the description of skin texture and irregular details, for which conventional vector representation are not suitable, brings forth the need of developing novel representations. To address the issue, Discriminative Multiscale Texton Features and SIFT-Activated Pictorial Structure are proposed to describe skin and subtle details respectively. To effectively combine the information conveyed by all layers, we further design an metric fusion algorithm adaptively placing emphasis onto the highly confident layers. Through systematic experiments, we identify different roles played by the layers and convincingly show that by utilizing their complementarities, our framework achieves remarkable performance improvement.

Inspired by the underlying relationship between classification capability and the mutual information, in this paper, we first establish a quantitative model to describe the information transmission process from feature extraction to final classification and identify the critical channel in this propagation path, and then propose a Maximum Effective Information Criteria for pursuing the optimal subspace in the sense of preserving maximum information that can be conveyed to final decision. Considering the orthogonality and rotation invariance properties of the solution space, we present a Conjugate Gradient method constrained on a Grassmann manifold to exploit the geometric traits of the solution space for enhancing the efficiency of optimization. Comprehensive experiments demonstrate that the framework integrating the Maximum Effective Information Criteria and Grassmann manifold-based optimization method significantly improves the classification performance.

Linear discriminant analysis (LDA) is a popular face recognition technique. However, an inherent problem with this technique stems from the parametric nature of the scatter matrix, in which the sample distribution in each class is assumed to be normal distribution. So it tends to suffer in the case of non-normal distribution. In this paper a nonparametric scatter matrix is defined to replace the traditional parametric scatter matrix in order to overcome this problem. Two kinds of nonparametric subspace analysis (NSA): PNSA and NNSA are proposed for face recognition. The former is based on the principal space of intra-personal scatter matrix, while the latter is based on the null space. In addition, based on the complementary nature of PNSA and NNSA, we further develop a dual NSA-based classifier framework using Gabor images to further improve the recognition performance. Experiments achieve near perfect recognition accuracy (99.7%) on the XM2VTS database.

In this paper, we propose a new face hallucination framework based on image patches, which integrates two novel statistical super-resolution models. Considering that image patches reflect the combined effect of personal characteristics and patch-location, we first formulate a TensorPatch model based on multilinear analysis to explicitly model the interaction between multiple constituent factors. Motivated by locally linear embedding, we develop an enhanced multilinear patch hallucination algorithm, which efficiently exploits the local distribution structure in the sample space. To better preserve face subtle details, we derive the coupled PCA algorithm to learn the relation between high-resolution residue and low-resolution residue, which is utilized for compensate the error residue in hallucinated images. Experiments demonstrate that our framework on one hand well maintains the global facial structures, on the other hand recovers the detailed facial traits in high quality.

In this paper, we present a new learning framework for image style transforms. Considering that the images in different style representations constitute different vector spaces, we propose a novel framework called coupled space learning to learn the relations between different spaces and use them to infer the images from one style to another style. Observing that for each style, only the components correlated to the space of the target style are useful for inference, we first develop the correlative component analysis to pursue the embedded hidden subspaces that best preserve the inter-space correlation information. Then we develop the coupled bidirectional transform algorithm to estimate the transforms between the two embedded spaces, where the coupling between the forward transform and the backward transform is explicitly taken into account. To enhance the capability of modelling complex data, we further develop the coupled Gaussian mixture model to generalize our framework to a mixture-model architecture. The effectiveness of the framework is demonstrated in the applications including face super-resolution and bidirectional portrait style transforms.

In this paper, we propose a novel framework for face super-resolution based on a layered predictor network. In the first layer, multiple predictors are trained online with a dynamic-constructed training set, which is adaptively selected in order to make the trained model tailored to the testing face. When the dynamic training set is obtained, the optimum predictor can be learned based on the resampling-maximum likelihood-model. To further enhance the robustness of prediction and the smoothness of the hallucinated image, additional layers are designed to fuse multiple predictors with the fusion rule learned from the training set. Experiments fully demonstrate the effectiveness of the framework.

Lighting condition is an important factor in face analysis and synthesis, which has received extensive study in both computer vision and computer graphics. Motivated by the work on multilinear model, we propose a learning-based algorithm for relighting based on tensor framework, which explicitly accounts for the interaction of the identity factor and the lighting factor. The major contribution of our work is that we develop a novel algorithm based on a two-stage decomposition scheme to simultaneously and robustly solve for the identity parameter and the lighting parameter which are both unknown. Equipped with the decomposition algorithm, the capability of the tensor model is significantly extended. Experiment results illustrate the effectiveness of our algorithm.

Linear discriminant analysis (LDA) is widely-used in face recognition systems. However, with the traditional formulation, the available information in the training samples is not sufficiently utilized. In this paper, we present a new formulation, called generalized LDA, where the scatter matrices are de defined in a more flexible manner by identifying the fundamental principles of the scatter matrices construction. We further propose a novel framework called feedback-based dynamic generalized LDA. It integrates the generalized LDA and the dynamic feedback strategy for subspace analysis, in which the subspace is iteratively optimized by utilizing the feedback from the previous step. The comparative experiments demonstrate that the new framework achieves encouraging improvement on performances of both the face identification and the face verification.

Recently many automatic face recognition (AFR) systems were developed for applications with unspecific persons, which is different from conventional pattern recognition problems where all classes are known in the training stage. In this paper, we present a systematic and comprehensive study on linear subspace methods for face recognition on unspecific persons. Over 6700 experiments using different algorithms with different training parameters and testing conditions are conducted on a large scale database (4550 samples) to investigate the compound effect of various influential factors. The observations based on these experiments are expected to provide widely applicable guidelines for designing practical AFR systems.

In this paper, we propose a novel patch-based face hallucination framework, which employs a dual model to hallucinate different components associated with one facial image. Our model is based on a statistical learning approach: associative learning. It suffices to learn the dependencies between low-resolution image patches and their high-resolution ones with a new concept hidden parameter space as a bridge to connect those patches with different resolutions. To compensate higher frequency information of images, we present a dual associative learning algorithm for orderly inferring main components and high frequency components of faces. The patches can be finally integrated to form a whole high-resolution image. Experiments demonstrate that our approach does render high quality superresolution faces.

In this paper, we propose a novel face hallucination framework based on image patches, which exploits local geometry structures of overlapping patches to hallucinate different components associated with one facial image. To achieve local fidelity while preserving smoothness in the target high-resolution image, we develop a neighbor combination super-resolution model for high-resolution patch synthesis. For further enhancing the detailed information, we propose another model, which effectively learns neighbor transformations between low- and high-resolution image patch residuals to compensate modeling errors caused by the first model. Experiments demonstrate that our approach can hallucinate high quality super-resolution faces.