Hezhen Hu

Zhendong Wang, Jianmin Bao, Wengang Zhou et al.

Diffusion models have shown remarkable success in visual synthesis, but have also raised concerns about potential abuse for malicious purposes. In this paper, we seek to build a detector for telling apart real images from diffusion-generated images. We find that existing detectors struggle to detect images generated by diffusion models, even if we include generated images from a specific diffusion model in their training data. To address this issue, we propose a novel image representation called DIffusion Reconstruction Error (DIRE), which measures the error between an input image and its reconstruction counterpart by a pre-trained diffusion model. We observe that diffusion-generated images can be approximately reconstructed by a diffusion model while real images cannot. It provides a hint that DIRE can serve as a bridge to distinguish generated and real images. DIRE provides an effective way to detect images generated by most diffusion models, and it is general for detecting generated images from unseen diffusion models and robust to various perturbations. Furthermore, we establish a comprehensive diffusion-generated benchmark including images generated by eight diffusion models to evaluate the performance of diffusion-generated image detectors. Extensive experiments on our collected benchmark demonstrate that DIRE exhibits superiority over previous generated-image detectors. The code and dataset are available at https://github.com/ZhendongWang6/DIRE.

Weichao Zhao, Hezhen Hu, Wengang Zhou et al. · tsinghua

In this work, we are dedicated to leveraging the BERT pre-training success and modeling the domain-specific statistics to fertilize the sign language recognition~(SLR) model. Considering the dominance of hand and body in sign language expression, we organize them as pose triplet units and feed them into the Transformer backbone in a frame-wise manner. Pre-training is performed via reconstructing the masked triplet unit from the corrupted input sequence, which learns the hierarchical correlation context cues among internal and external triplet units. Notably, different from the highly semantic word token in BERT, the pose unit is a low-level signal originally located in continuous space, which prevents the direct adoption of the BERT cross-entropy objective. To this end, we bridge this semantic gap via coupling tokenization of the triplet unit. It adaptively extracts the discrete pseudo label from the pose triplet unit, which represents the semantic gesture/body state. After pre-training, we fine-tune the pre-trained encoder on the downstream SLR task, jointly with the newly added task-specific layer. Extensive experiments are conducted to validate the effectiveness of our proposed method, achieving new state-of-the-art performance on all four benchmarks with a notable gain.

Hezhen Hu, Wangbo Zhao, Lanqing Guo et al.

In this paper, we present HumanNOVA, a photorealistic, universal, and rapid model for generating 3D human avatars from a single RGB image. Achieving both photorealism and generalization is challenging due to the scarcity of diverse, high-quality 3D human data. To address this, we build a scalable data generation pipeline that follows two strategies. The first one is to leverage existing rigged assets and animate them with extensive poses from daily life. The second strategy is to utilize existing multi-camera captures of humans and employ fitting to generate more diverse views for training. These two strategies enable us to scale up to 100k assets, significantly enhancing both the quantity and the diversity of data for robust model training. In terms of the architecture, HumanNOVA adopts a feed-forward, token-conditioned avatar modeling framework that allows fast inference in less than one second and requires no test-time optimization. Given an input image and an estimated simplified human mesh (SMPL) without detailed geometry or appearance, the model first encodes both inputs into compact token representations. These tokens then act as conditioning signals and are fused through cross-attention to construct a triplane-based 3D avatar representation. Extensive experiments on multiple benchmarks demonstrate the superiority of our approach, both quantitatively and qualitatively, as well as its robustness under diverse input image conditions. Project page at https://HumanNOVA.github.io .

Huijie Yao, Wengang Zhou, Hao Feng et al.

This paper presents IP-SLT, a simple yet effective framework for sign language translation (SLT). Our IP-SLT adopts a recurrent structure and enhances the semantic representation (prototype) of the input sign language video via an iterative refinement manner. Our idea mimics the behavior of human reading, where a sentence can be digested repeatedly, till reaching accurate understanding. Technically, IP-SLT consists of feature extraction, prototype initialization, and iterative prototype refinement. The initialization module generates the initial prototype based on the visual feature extracted by the feature extraction module. Then, the iterative refinement module leverages the cross-attention mechanism to polish the previous prototype by aggregating it with the original video feature. Through repeated refinement, the prototype finally converges to a more stable and accurate state, leading to a fluent and appropriate translation. In addition, to leverage the sequential dependence of prototypes, we further propose an iterative distillation loss to compress the knowledge of the final iteration into previous ones. As the autoregressive decoding process is executed only once in inference, our IP-SLT is ready to improve various SLT systems with acceptable overhead. Extensive experiments are conducted on public benchmarks to demonstrate the effectiveness of the IP-SLT.

Hezhen Hu, Weilun Wang, Wengang Zhou et al.

In this work, we are dedicated to a new task, i.e., hand-object interaction image generation, which aims to conditionally generate the hand-object image under the given hand, object and their interaction status. This task is challenging and research-worthy in many potential application scenarios, such as AR/VR games and online shopping, etc. To address this problem, we propose a novel HOGAN framework, which utilizes the expressive model-aware hand-object representation and leverages its inherent topology to build the unified surface space. In this space, we explicitly consider the complex self- and mutual occlusion during interaction. During final image synthesis, we consider different characteristics of hand and object and generate the target image in a split-and-combine manner. For evaluation, we build a comprehensive protocol to access both the fidelity and structure preservation of the generated image. Extensive experiments on two large-scale datasets, i.e., HO3Dv3 and DexYCB, demonstrate the effectiveness and superiority of our framework both quantitatively and qualitatively. The project page is available at https://play-with-hoi-generation.github.io/.

Wengang Zhou, Weichao Zhao, Hezhen Hu et al.

Sign language serves as the primary meaning of communication for the deaf-mute community. Different from spoken language, it commonly conveys information by the collaboration of manual features, i.e., hand gestures and body movements, and non-manual features, i.e., facial expressions and mouth cues. To facilitate communication between the deaf-mute and hearing people, a series of sign language understanding (SLU) tasks have been studied in recent years, including isolated/continuous sign language recognition (ISLR/CSLR), gloss-free sign language translation (GF-SLT) and sign language retrieval (SL-RT). Sign language recognition and translation aims to understand the semantic meaning conveyed by sign languages from gloss-level and sentence-level, respectively. In contrast, SL-RT focuses on retrieving sign videos or corresponding texts from a closed-set under the query-by-example search paradigm. These tasks investigate sign language topics from diverse perspectives and raise challenges in learning effective representation of sign language videos. To advance the development of sign language understanding, exploring a generalized model that is applicable across various SLU tasks is a profound research direction.

Weichao Zhao, Hezhen Hu, Wengang Zhou et al.

Reconstructing interacting hands from monocular RGB data is a challenging task, as it involves many interfering factors, e.g. self- and mutual occlusion and similar textures. Previous works only leverage information from a single RGB image without modeling their physically plausible relation, which leads to inferior reconstruction results. In this work, we are dedicated to explicitly exploiting spatial-temporal information to achieve better interacting hand reconstruction. On one hand, we leverage temporal context to complement insufficient information provided by the single frame, and design a novel temporal framework with a temporal constraint for interacting hand motion smoothness. On the other hand, we further propose an interpenetration detection module to produce kinetically plausible interacting hands without physical collisions. Extensive experiments are performed to validate the effectiveness of our proposed framework, which achieves new state-of-the-art performance on public benchmarks.

Hezhen Hu, Xiaoyi Dong, Jianmin Bao et al.

Pre-training is playing an increasingly important role in learning generic feature representation for Person Re-identification (ReID). We argue that a high-quality ReID representation should have three properties, namely, multi-level awareness, occlusion robustness, and cross-region invariance. To this end, we propose a simple yet effective pre-training framework, namely PersonMAE, which involves two core designs into masked autoencoders to better serve the task of Person Re-ID. 1) PersonMAE generates two regions from the given image with RegionA as the input and \textit{RegionB} as the prediction target. RegionA is corrupted with block-wise masking to mimic common occlusion in ReID and its remaining visible parts are fed into the encoder. 2) Then PersonMAE aims to predict the whole RegionB at both pixel level and semantic feature level. It encourages its pre-trained feature representations with the three properties mentioned above. These properties make PersonMAE compatible with downstream Person ReID tasks, leading to state-of-the-art performance on four downstream ReID tasks, i.e., supervised (holistic and occluded setting), and unsupervised (UDA and USL setting). Notably, on the commonly adopted supervised setting, PersonMAE with ViT-B backbone achieves 79.8% and 69.5% mAP on the MSMT17 and OccDuke datasets, surpassing the previous state-of-the-art by a large margin of +8.0 mAP, and +5.3 mAP, respectively.

Zhiwen Fan, Jian Zhang, Renjie Li et al.

The rapid advancement of Large Multimodal Models (LMMs) for 2D images and videos has motivated extending these models to understand 3D scenes, aiming for human-like visual-spatial intelligence. Nevertheless, achieving deep spatial understanding comparable to human capabilities poses significant challenges in model encoding and data acquisition. Existing methods frequently depend on external depth sensors for geometry capture or utilize off-the-shelf algorithms for pre-constructing 3D maps, thereby limiting their scalability, especially with prevalent monocular video inputs and for time-sensitive applications. In this work, we introduce VLM-3R, a unified framework for Vision-Language Models (VLMs) that incorporates 3D Reconstructive instruction tuning. VLM-3R processes monocular video frames by employing a geometry encoder to derive implicit 3D tokens that represent spatial understanding. Leveraging our Spatial-Visual-View Fusion and over 200K curated 3D reconstructive instruction tuning question-answer (QA) pairs, VLM-3R effectively aligns real-world spatial context with language instructions. This enables monocular 3D spatial assistance and embodied reasoning. To facilitate the evaluation of temporal reasoning, we introduce the Vision-Spatial-Temporal Intelligence benchmark, featuring over 138.6K QA pairs across five distinct tasks focused on evolving spatial relationships. Extensive experiments demonstrate that our model, VLM-3R, not only facilitates robust visual-spatial reasoning but also enables the understanding of temporal 3D context changes, excelling in both accuracy and scalability.

Hezhen Hu, Zhiwen Fan, Tianhao Wu et al.

Nuanced expressiveness, particularly through fine-grained hand and facial expressions, is pivotal for enhancing the realism and vitality of digital human representations. In this work, we focus on investigating the expressiveness of human avatars when learned from monocular RGB video; a setting that introduces new challenges in capturing and animating fine-grained details. To this end, we introduce EVA, a drivable human model that meticulously sculpts fine details based on 3D Gaussians and SMPL-X, an expressive parametric human model. Focused on enhancing expressiveness, our work makes three key contributions. First, we highlight the critical importance of aligning the SMPL-X model with RGB frames for effective avatar learning. Recognizing the limitations of current SMPL-X prediction methods for in-the-wild videos, we introduce a plug-and-play module that significantly ameliorates misalignment issues. Second, we propose a context-aware adaptive density control strategy, which is adaptively adjusting the gradient thresholds to accommodate the varied granularity across body parts. Last but not least, we develop a feedback mechanism that predicts per-pixel confidence to better guide the learning of 3D Gaussians. Extensive experiments on two benchmarks demonstrate the superiority of our framework both quantitatively and qualitatively, especially on the fine-grained hand and facial details. See the project website at \url{https://evahuman.github.io}

Zecheng Li, Wengang Zhou, Weichao Zhao et al.

Sign language pre-training has gained increasing attention for its ability to enhance performance across various sign language understanding (SLU) tasks. However, existing methods often suffer from a gap between pre-training and fine-tuning, leading to suboptimal results. To address this, we propose Uni-Sign, a unified pre-training framework that eliminates the gap between pre-training and downstream SLU tasks through a large-scale generative pre-training strategy and a novel fine-tuning paradigm. First, we introduce CSL-News, a large-scale Chinese Sign Language (CSL) dataset containing 1,985 hours of video paired with textual annotations, which enables effective large-scale pre-training. Second, Uni-Sign unifies SLU tasks by treating downstream tasks as a single sign language translation (SLT) task during fine-tuning, ensuring seamless knowledge transfer between pre-training and fine-tuning. Furthermore, we incorporate a prior-guided fusion (PGF) module and a score-aware sampling strategy to efficiently fuse pose and RGB information, addressing keypoint inaccuracies and improving computational efficiency. Extensive experiments across multiple SLU benchmarks demonstrate that Uni-Sign achieves state-of-the-art performance across multiple downstream SLU tasks. Dataset and code are available at github.com/ZechengLi19/Uni-Sign.

Weichao Zhao, Wengang Zhou, Hezhen Hu et al.

Recently, there have been efforts to improve the performance in sign language recognition by designing self-supervised learning methods. However, these methods capture limited information from sign pose data in a frame-wise learning manner, leading to sub-optimal solutions. To this end, we propose a simple yet effective self-supervised contrastive learning framework to excavate rich context via spatial-temporal consistency from two distinct perspectives and learn instance discriminative representation for sign language recognition. On one hand, since the semantics of sign language are expressed by the cooperation of fine-grained hands and coarse-grained trunks, we utilize both granularity information and encode them into latent spaces. The consistency between hand and trunk features is constrained to encourage learning consistent representation of instance samples. On the other hand, inspired by the complementary property of motion and joint modalities, we first introduce first-order motion information into sign language modeling. Additionally, we further bridge the interaction between the embedding spaces of both modalities, facilitating bidirectional knowledge transfer to enhance sign language representation. Our method is evaluated with extensive experiments on four public benchmarks, and achieves new state-of-the-art performance with a notable margin. The source code is publicly available at https://github.com/sakura/Code.

Dejia Xu, Hanwen Liang, Neel P. Bhatt et al.

Recent advancements in diffusion models for 2D and 3D content creation have sparked a surge of interest in generating 4D content. However, the scarcity of 3D scene datasets constrains current methodologies to primarily object-centric generation. To overcome this limitation, we present Comp4D, a novel framework for Compositional 4D Generation. Unlike conventional methods that generate a singular 4D representation of the entire scene, Comp4D innovatively constructs each 4D object within the scene separately. Utilizing Large Language Models (LLMs), the framework begins by decomposing an input text prompt into distinct entities and maps out their trajectories. It then constructs the compositional 4D scene by accurately positioning these objects along their designated paths. To refine the scene, our method employs a compositional score distillation technique guided by the pre-defined trajectories, utilizing pre-trained diffusion models across text-to-image, text-to-video, and text-to-3D domains. Extensive experiments demonstrate our outstanding 4D content creation capability compared to prior arts, showcasing superior visual quality, motion fidelity, and enhanced object interactions.

Renjie Li, Ruijie Ye, Mingyang Wu et al.

Humans are integral components of the transportation ecosystem, and understanding their behaviors is crucial to facilitating the development of safe driving systems. Although recent progress has explored various aspects of human behavior$\unicode{x2014}$such as motion, trajectories, and intention$\unicode{x2014}$a comprehensive benchmark for evaluating human behavior understanding in autonomous driving remains unavailable. In this work, we propose $\textbf{MMHU}$, a large-scale benchmark for human behavior analysis featuring rich annotations, such as human motion and trajectories, text description for human motions, human intention, and critical behavior labels relevant to driving safety. Our dataset encompasses 57k human motion clips and 1.73M frames gathered from diverse sources, including established driving datasets such as Waymo, in-the-wild videos from YouTube, and self-collected data. A human-in-the-loop annotation pipeline is developed to generate rich behavior captions. We provide a thorough dataset analysis and benchmark multiple tasks$\unicode{x2014}$ranging from motion prediction to motion generation and human behavior question answering$\unicode{x2014}$thereby offering a broad evaluation suite. Project page : https://MMHU-Benchmark.github.io.

Lulin Liu, Dayou Li, Yiqing Liang et al.

Large foundation models have made significant advances in embodied intelligence, enabling synthesis and reasoning over egocentric input for household tasks. However, VLM-based auto-labeling is often noisy because the primary data sources lack accurate human action labels, chain-of-thought (CoT), and spatial annotations; these errors are amplified during long-horizon spatial instruction following. These issues stem from insufficient coverage of minute-long, daily household planning tasks and from inaccurate spatial grounding. As a result, VLM reasoning chains and world-model synthesis can hallucinate objects, skip steps, or fail to respect real-world physical attributes. To address these gaps, we introduce EgoTL. EgoTL builds a think-aloud capture pipeline for egocentric data. It uses a say-before-act protocol to record step-by-step goals and spoken reasoning with word-level timestamps, then calibrates physical properties with metric-scale spatial estimators, a memory-bank walkthrough for scene context, and clip-level tags for navigation instructions and detailed manipulation actions. With EgoTL, we are able to benchmark VLMs and World Models on six task dimensions from three layers and long-horizon generation over minute-long sequences across over 100 daily household tasks. We find that foundation models still fall short as egocentric assistants or open-world simulators. Finally, we finetune foundation models with human CoT aligned with metric labels on the training split of EgoTL, which improves long-horizon planning and reasoning, step-wise reasoning, instruction following, and spatial grounding.

Kepeng Wu, Zecheng Li, Hezhen Hu et al.

Pre-training has been proven to be effective in boosting the performance of Isolated Sign Language Recognition (ISLR). Existing pre-training methods solely focus on the compact pose data, which eliminates background perturbation but inevitably suffers from insufficient semantic cues compared to raw RGB videos. Nevertheless, learning representation directly from RGB videos remains challenging due to the presence of sign-independent visual features. To address this dilemma, we propose a Cross-modal Consistency Learning framework (CCL-SLR), which leverages the cross-modal consistency from both RGB and pose modalities based on self-supervised pre-training. First, CCL-SLR employs contrastive learning for instance discrimination within and across modalities. Through the single-modal and cross-modal contrastive learning, CCL-SLR gradually aligns the feature spaces of RGB and pose modalities, thereby extracting consistent sign representations. Second, we further introduce Motion-Preserving Masking (MPM) and Semantic Positive Mining (SPM) techniques to improve cross-modal consistency from the perspective of data augmentation and sample similarity, respectively. Extensive experiments on four ISLR benchmarks show that CCL-SLR achieves impressive performance, demonstrating its effectiveness. The code will be released to the public.

Lanqing Guo, Yufei Wang, Hezhen Hu et al.

Many 3D scene editing tasks focus on modifying local regions rather than the entire scene, except for some global applications like style transfer, and in the context of 3D Gaussian Splatting (3DGS), where scenes are represented by a series of Gaussians, this structure allows for precise regional edits, offering enhanced control over specific areas of the scene; however, the challenge lies in the fact that 3D semantic parsing often underperforms compared to its 2D counterpart, making targeted manipulations within 3D spaces more difficult and limiting the fidelity of edits, which we address by leveraging 2D diffusion editing to accurately identify modification regions in each view, followed by inverse rendering for 3D localization, then refining the frontal view and initializing a coarse 3DGS with consistent views and approximate shapes derived from depth maps predicted by a 2D foundation model, thereby supporting an iterative, view-consistent editing process that gradually enhances structural details and textures to ensure coherence across perspectives. Experiments demonstrate that our method achieves state-of-the-art performance while delivering up to a $4\times$ speedup, providing a more efficient and effective approach to 3D scene local editing.

Hezhen Hu, Weichao Zhao, Wengang Zhou et al.

Hand gesture serves as a crucial role during the expression of sign language. Current deep learning based methods for sign language understanding (SLU) are prone to over-fitting due to insufficient sign data resource and suffer limited interpretability. In this paper, we propose the first self-supervised pre-trainable SignBERT+ framework with model-aware hand prior incorporated. In our framework, the hand pose is regarded as a visual token, which is derived from an off-the-shelf detector. Each visual token is embedded with gesture state and spatial-temporal position encoding. To take full advantage of current sign data resource, we first perform self-supervised learning to model its statistics. To this end, we design multi-level masked modeling strategies (joint, frame and clip) to mimic common failure detection cases. Jointly with these masked modeling strategies, we incorporate model-aware hand prior to better capture hierarchical context over the sequence. After the pre-training, we carefully design simple yet effective prediction heads for downstream tasks. To validate the effectiveness of our framework, we perform extensive experiments on three main SLU tasks, involving isolated and continuous sign language recognition (SLR), and sign language translation (SLT). Experimental results demonstrate the effectiveness of our method, achieving new state-of-the-art performance with a notable gain.

Hezhen Hu, Weichao Zhao, Wengang Zhou et al.

Hand gesture serves as a critical role in sign language. Current deep-learning-based sign language recognition (SLR) methods may suffer insufficient interpretability and overfitting due to limited sign data sources. In this paper, we introduce the first self-supervised pre-trainable SignBERT with incorporated hand prior for SLR. SignBERT views the hand pose as a visual token, which is derived from an off-the-shelf pose extractor. The visual tokens are then embedded with gesture state, temporal and hand chirality information. To take full advantage of available sign data sources, SignBERT first performs self-supervised pre-training by masking and reconstructing visual tokens. Jointly with several mask modeling strategies, we attempt to incorporate hand prior in a model-aware method to better model hierarchical context over the hand sequence. Then with the prediction head added, SignBERT is fine-tuned to perform the downstream SLR task. To validate the effectiveness of our method on SLR, we perform extensive experiments on four public benchmark datasets, i.e., NMFs-CSL, SLR500, MSASL and WLASL. Experiment results demonstrate the effectiveness of both self-supervised learning and imported hand prior. Furthermore, we achieve state-of-the-art performance on all benchmarks with a notable gain.

Junfu Pu, Wengang Zhou, Hezhen Hu et al.

Continuous sign language recognition (SLR) deals with unaligned video-text pair and uses the word error rate (WER), i.e., edit distance, as the main evaluation metric. Since it is not differentiable, we usually instead optimize the learning model with the connectionist temporal classification (CTC) objective loss, which maximizes the posterior probability over the sequential alignment. Due to the optimization gap, the predicted sentence with the highest decoding probability may not be the best choice under the WER metric. To tackle this issue, we propose a novel architecture with cross modality augmentation. Specifically, we first augment cross-modal data by simulating the calculation procedure of WER, i.e., substitution, deletion and insertion on both text label and its corresponding video. With these real and generated pseudo video-text pairs, we propose multiple loss terms to minimize the cross modality distance between the video and ground truth label, and make the network distinguish the difference between real and pseudo modalities. The proposed framework can be easily extended to other existing CTC based continuous SLR architectures. Extensive experiments on two continuous SLR benchmarks, i.e., RWTH-PHOENIX-Weather and CSL, validate the effectiveness of our proposed method.

Hezhen Hu, Wengang Zhou, Junfu Pu et al.

Sign language recognition (SLR) is a challenging problem, involving complex manual features, i.e., hand gestures, and fine-grained non-manual features (NMFs), i.e., facial expression, mouth shapes, etc. Although manual features are dominant, non-manual features also play an important role in the expression of a sign word. Specifically, many sign words convey different meanings due to non-manual features, even though they share the same hand gestures. This ambiguity introduces great challenges in the recognition of sign words. To tackle the above issue, we propose a simple yet effective architecture called Global-local Enhancement Network (GLE-Net), including two mutually promoted streams towards different crucial aspects of SLR. Of the two streams, one captures the global contextual relationship, while the other stream captures the discriminative fine-grained cues. Moreover, due to the lack of datasets explicitly focusing on this kind of features, we introduce the first non-manual-features-aware isolated Chinese sign language dataset~(NMFs-CSL) with a total vocabulary size of 1,067 sign words in daily life. Extensive experiments on NMFs-CSL and SLR500 datasets demonstrate the effectiveness of our method.