Fangrui Zhu

Fangrui Zhu, Yiming Xie, Weidi Xie et al.

We have witnessed significant progress in human-object interaction (HOI) detection. The reliance on mAP (mean Average Precision) scores as a summary metric, however, does not provide sufficient insight into the nuances of model performance (e.g., why one model is better than another), which can hinder further innovation in this field. To address this issue, in this paper, we introduce a diagnosis toolbox to provide detailed quantitative break-down analysis of HOI detection models, inspired by the success of object detection diagnosis toolboxes. We first conduct holistic investigations in the pipeline of HOI detection. By defining a set of errors and the oracles to fix each of them, we can have a quantitative analysis of the significance of different errors according to the mAP improvement obtained from fixing each error. We then delve into two sub-tasks of HOI detection: human-object pair detection and interaction classification, respectively. For the first detection task, we compute the coverage of ground-truth human-object pairs as well as the noisiness level in the detection results. For the second classification task, we measure a model's performance of differentiating positive and negative detection results and also classifying the actual interactions when the human-object pairs are correctly detected. We analyze eight state-of-the-art HOI detection models and provide valuable diagnosis insights to foster future research. For instance, our diagnosis shows that state-of-the-art model RLIPv2 outperforms others mainly because it significantly improves the multi-label interaction classification accuracy. Our toolbox is applicable for different methods across different datasets and available at https://github.com/neu-vi/Diag-HOI.

Zeyu Han, Fangrui Zhu, Qianru Lao et al.

Zero-shot referring expression comprehension aims at localizing bounding boxes in an image corresponding to provided textual prompts, which requires: (i) a fine-grained disentanglement of complex visual scene and textual context, and (ii) a capacity to understand relationships among disentangled entities. Unfortunately, existing large vision-language alignment (VLA) models, e.g., CLIP, struggle with both aspects so cannot be directly used for this task. To mitigate this gap, we leverage large foundation models to disentangle both images and texts into triplets in the format of (subject, predicate, object). After that, grounding is accomplished by calculating the structural similarity matrix between visual and textual triplets with a VLA model, and subsequently propagate it to an instance-level similarity matrix. Furthermore, to equip VLA models with the ability of relationship understanding, we design a triplet-matching objective to fine-tune the VLA models on a collection of curated dataset containing abundant entity relationships. Experiments demonstrate that our visual grounding performance increase of up to 19.5% over the SOTA zero-shot model on RefCOCO/+/g. On the more challenging Who's Waldo dataset, our zero-shot approach achieves comparable accuracy to the fully supervised model. Code is available at https://github.com/Show-han/Zeroshot_REC.

Jing Gu, Xian Liu, Yu Zeng et al.

Video generation models have achieved remarkable progress in creating high-quality, photorealistic content. However, their ability to accurately simulate physical phenomena remains a critical and unresolved challenge. This paper presents PhyWorldBench, a comprehensive benchmark designed to evaluate video generation models based on their adherence to the laws of physics. The benchmark covers multiple levels of physical phenomena, ranging from fundamental principles like object motion and energy conservation to more complex scenarios involving rigid body interactions and human or animal motion. Additionally, we introduce a novel ""Anti-Physics"" category, where prompts intentionally violate real-world physics, enabling the assessment of whether models can follow such instructions while maintaining logical consistency. Besides large-scale human evaluation, we also design a simple yet effective method that could utilize current MLLM to evaluate the physics realism in a zero-shot fashion. We evaluate 12 state-of-the-art text-to-video generation models, including five open-source and five proprietary models, with a detailed comparison and analysis. we identify pivotal challenges models face in adhering to real-world physics. Through systematic testing of their outputs across 1,050 curated prompts-spanning fundamental, composite, and anti-physics scenarios-we identify pivotal challenges these models face in adhering to real-world physics. We then rigorously examine their performance on diverse physical phenomena with varying prompt types, deriving targeted recommendations for crafting prompts that enhance fidelity to physical principles.

Fangrui Zhu, Jianwei Yang, Huaizu Jiang

Visual relationship understanding has been studied separately in human-object interaction(HOI) detection, scene graph generation(SGG), and referring relationships(RR) tasks. Given the complexity and interconnectedness of these tasks, it is crucial to have a flexible framework that can effectively address these tasks in a cohesive manner. In this work, we propose FleVRS, a single model that seamlessly integrates the above three aspects in standard and promptable visual relationship segmentation, and further possesses the capability for open-vocabulary segmentation to adapt to novel scenarios. FleVRS leverages the synergy between text and image modalities, to ground various types of relationships from images and use textual features from vision-language models to visual conceptual understanding. Empirical validation across various datasets demonstrates that our framework outperforms existing models in standard, promptable, and open-vocabulary tasks, e.g., +1.9 $mAP$ on HICO-DET, +11.4 $Acc$ on VRD, +4.7 $mAP$ on unseen HICO-DET. Our FleVRS represents a significant step towards a more intuitive, comprehensive, and scalable understanding of visual relationships.

Fangrui Zhu, Hanhui Wang, Yiming Xie et al.

Unlocking spatial reasoning in Multimodal Large Language Models (MLLMs) is crucial for enabling intelligent interaction with 3D environments. While prior efforts often rely on explicit 3D inputs or specialized model architectures, we ask: can MLLMs reason about 3D space using only structured 2D representations derived from perception? We introduce Struct2D, a perception-guided prompting framework that combines bird's-eye-view (BEV) images with object marks and object-centric metadata, optionally incorporating egocentric keyframes when needed. Using Struct2D, we conduct an in-depth zero-shot analysis of closed-source MLLMs (e.g., GPT-o3) and find that they exhibit surprisingly strong spatial reasoning abilities when provided with structured 2D inputs, effectively handling tasks such as relative direction estimation and route planning. Building on these insights, we construct Struct2D-Set, a large-scale instruction tuning dataset with 200K fine-grained QA pairs across eight spatial reasoning categories, generated automatically from 3D indoor scenes. We fine-tune an open-source MLLM (Qwen2.5VL) on Struct2D-Set, achieving competitive performance on multiple benchmarks, including 3D question answering, dense captioning, and object grounding. Our approach demonstrates that structured 2D inputs can effectively bridge perception and language reasoning in MLLMs-without requiring explicit 3D representations as input. We will release both our code and dataset to support future research.

Fangrui Zhu, Li Zhang, Yanwei Fu et al.

The objective of this paper is self-supervised representation learning, with the goal of solving semi-supervised video object segmentation (a.k.a. dense tracking). We make the following contributions: (i) we propose to improve the existing self-supervised approach, with a simple, yet more effective memory mechanism for long-term correspondence matching, which resolves the challenge caused by the dis-appearance and reappearance of objects; (ii) by augmenting the self-supervised approach with an online adaptation module, our method successfully alleviates tracker drifts caused by spatial-temporal discontinuity, e.g. occlusions or dis-occlusions, fast motions; (iii) we explore the efficiency of self-supervised representation learning for dense tracking, surprisingly, we show that a powerful tracking model can be trained with as few as 100 raw video clips (equivalent to a duration of 11mins), indicating that low-level statistics have already been effective for tracking tasks; (iv) we demonstrate state-of-the-art results among the self-supervised approaches on DAVIS-2017 and YouTube-VOS, as well as surpassing most of methods trained with millions of manual segmentation annotations, further bridging the gap between self-supervised and supervised learning. Codes are released to foster any further research (https://github.com/fangruizhu/self_sup_semiVOS).

Mu Cai, Reuben Tan, Jianrui Zhang et al.

Understanding fine-grained temporal dynamics is crucial for multimodal video comprehension and generation. Due to the lack of fine-grained temporal annotations, existing video benchmarks mostly resemble static image benchmarks and are incompetent at evaluating models for temporal understanding. In this paper, we introduce TemporalBench, a new benchmark dedicated to evaluating fine-grained temporal understanding in videos. TemporalBench consists of ~10K video question-answer pairs, derived from ~2K high-quality human annotations detailing the temporal dynamics in video clips. As a result, our benchmark provides a unique testbed for evaluating various temporal understanding and reasoning abilities such as action frequency, motion magnitude, event order, etc. Moreover, it enables evaluations on various tasks like both video question answering and captioning, both short and long video understanding, as well as different models such as multimodal video embedding models and text generation models. Results show that state-of-the-art models like GPT-4o achieve only 38.5% question answering accuracy on TemporalBench, demonstrating a significant gap (~30%) between humans and AI in temporal understanding. Furthermore, we notice a critical pitfall for multi-choice QA where LLMs can detect the subtle changes in negative captions and find a centralized description as a cue for its prediction, where we propose Multiple Binary Accuracy (MBA) to correct such bias. We hope that TemporalBench can foster research on improving models' temporal reasoning capabilities. Both dataset and evaluation code will be made available.

Fangrui Zhu, Yunfeng Xi, Jianmo Ni et al.

Egocentric video understanding is inherently complex due to the dynamic 4D nature of the environment, where camera motion and object displacements necessitate a continuous re-evaluation of spatial relations. In this work, we target a suite of under-explored egocentric 4D reasoning tasks, including fixture interaction counting, viewpoint-relative fixture location, object movement itinerary tracking, and stationary object localization, that require fundamentally different cognitive operations: spatial anchoring, temporal tracking, and duration reasoning. We observe that these structural differences make task-agnostic approaches insufficient: generic Chain-of-Thought methods lack task-appropriate reasoning primitives, and uniform reinforcement learning actively destabilizes performance on spatial tasks. To address this, we propose EgoReasoner, a two-stage framework that aligns both the reasoning scaffold and the reward signal to each task's cognitive structure. In the first stage, Task-Adaptive Thinking Templates guide the synthesis of structured CoT traces that teach the model to reason adaptively across task types via supervised fine-tuning. In the second stage, task-aware reward functions verify entity grounding, temporal alignment, and task-adaptive logical consistency, selectively strengthening each reasoning pathway via reinforcement fine-tuning with GRPO. Our 3B-parameter model, trained on only 16K samples, achieves 37.5% average accuracy on the challenging HD-EPIC benchmark, surpassing Qwen2.5-VL-7B (25.7%) by over 10 points.

Fangrui Zhu, Yi Zhu, Li Zhang et al.

Semantic segmentation is a challenging problem due to difficulties in modeling context in complex scenes and class confusions along boundaries. Most literature either focuses on context modeling or boundary refinement, which is less generalizable in open-world scenarios. In this work, we advocate a unified framework(UN-EPT) to segment objects by considering both context information and boundary artifacts. We first adapt a sparse sampling strategy to incorporate the transformer-based attention mechanism for efficient context modeling. In addition, a separate spatial branch is introduced to capture image details for boundary refinement. The whole model can be trained in an end-to-end manner. We demonstrate promising performance on three popular benchmarks for semantic segmentation with low memory footprint. Code will be released soon.

Xuelin Qian, Wenxuan Wang, Li Zhang et al.

Person re-identification (Re-ID) aims to match a target person across camera views at different locations and times. Existing Re-ID studies focus on the short-term cloth-consistent setting, under which a person re-appears in different camera views with the same outfit. A discriminative feature representation learned by existing deep Re-ID models is thus dominated by the visual appearance of clothing. In this work, we focus on a much more difficult yet practical setting where person matching is conducted over long-duration, e.g., over days and months and therefore inevitably under the new challenge of changing clothes. This problem, termed Long-Term Cloth-Changing (LTCC) Re-ID is much understudied due to the lack of large scale datasets. The first contribution of this work is a new LTCC dataset containing people captured over a long period of time with frequent clothing changes. As a second contribution, we propose a novel Re-ID method specifically designed to address the cloth-changing challenge. Specifically, we consider that under cloth-changes, soft-biometrics such as body shape would be more reliable. We, therefore, introduce a shape embedding module as well as a cloth-elimination shape-distillation module aiming to eliminate the now unreliable clothing appearance features and focus on the body shape information. Extensive experiments show that superior performance is achieved by the proposed model on the new LTCC dataset. The code and dataset will be available at https://naiq.github.io/LTCC_Perosn_ReID.html.

Jianhui Chen, Fangrui Zhu, James J. Little

Calibrating narrow field of view soccer cameras is challenging because there are very few field markings in the image. Unlike previous solutions, we propose a two-point method, which requires only two point correspondences given the prior knowledge of base location and orientation of a pan-tilt-zoom (PTZ) camera. We deploy this new calibration method to annotate pan-tilt-zoom data from soccer videos. The collected data are used as references for new images. We also propose a fast random forest method to predict pan-tilt angles without image-to-image feature matching, leading to an efficient calibration method for new images. We demonstrate our system on synthetic data and two real soccer datasets. Our two-point approach achieves superior performance over the state-of-the-art method.