Mengshi Qi

Wei Deng, Xianlin Zhang, Mengshi Qi

Enabling Vision-Language Models (VLMs) to perform spatial reasoning remains challenging. Existing approaches treat VLMs as passive observers, which is difficult for real-world applications. Moreover, reinforcement learning methods rely on sparse rewards, limiting their effectiveness for complex reasoning tasks. Inspired by pigeons' building and exploiting cognitive maps for navigation, we propose a novel agentic pipeline for spatial reasoning. First, we introduce a new \emph{dynamic cognitive map} parameterizing scene layout as object positions and orientations, serving as persistent memory for new observations. Second, we propose a novel \emph{Spatial Assertion Codes (SAC)}, Python expressions programmatically describing spatial relationships. By collaborating with the dynamic cognitive map, SAC enables verification of intermediate reasoning steps, providing dense reward signals. We optimize the model via supervised and reinforcement finetuning. Experiments on the MindCube benchmark demonstrate state-of-the-art performance with \emph{80.5\%} overall accuracy, outperforming the best current method by \emph{29.5} accuracy points (a relative improvement of \emph{53.2\%}) on the challenging \textsc{Rotation} subset. Our code and data are open-sourced at https://github.com/dw-dengwei/active-spatial-reasoning.git.

Changsheng Lv, Mengshi Qi, Xia Li et al.

In this paper, we propose a novel model called SGFormer, Semantic Graph TransFormer for point cloud-based 3D scene graph generation. The task aims to parse a point cloud-based scene into a semantic structural graph, with the core challenge of modeling the complex global structure. Existing methods based on graph convolutional networks (GCNs) suffer from the over-smoothing dilemma and can only propagate information from limited neighboring nodes. In contrast, SGFormer uses Transformer layers as the base building block to allow global information passing, with two types of newly-designed layers tailored for the 3D scene graph generation task. Specifically, we introduce the graph embedding layer to best utilize the global information in graph edges while maintaining comparable computation costs. Furthermore, we propose the semantic injection layer to leverage linguistic knowledge from large-scale language model (i.e., ChatGPT), to enhance objects' visual features. We benchmark our SGFormer on the established 3DSSG dataset and achieve a 40.94% absolute improvement in relationship prediction's R@50 and an 88.36% boost on the subset with complex scenes over the state-of-the-art. Our analyses further show SGFormer's superiority in the long-tail and zero-shot scenarios. Our source code is available at https://github.com/Andy20178/SGFormer.

Mengshi Qi, Wei Deng, Xianlin Zhang et al.

Large Vision-Language Models have achieved significant reasoning performance in various tasks.However, there are few studies on text-to-3D indoor scene generation with LVLMs. The main challenge is that prevailing LVLM-based methods employ chain-of-thought sequential decision mechanisms that cannot revise earlier decisions, causing error propagation.In this paper, we consider the task as a planning problem constrained by spatial and layout commonsense.To solve this problem, we model it as a tree search problem with global and local trees, which differs from existing sequential decision-making approaches.In the global tree, we place each object iteratively and explore multiple attempts like humans furnishing a room, where the problem space is represented as a tree.To effectively search the tree, we propose a hierarchical scene representation and a PRM-guided MCTS method.The hierarchical representation abstracts a scene into room level, region level, floor object level, and supported object level.The PRM-guided MCTS method uses the PRM to prune unnecessary branches and the MCTS algorithm to balance exploration and exploitation to get an optimal solution with fewer attempts.In the local tree, it further decomposes the placement of each object into finer sub-steps, including the specific placement parameters.To make the whole appearance of the scene consistent, we leverage pre-trained diffusion image generative models to predict textures for all the objects in the scene.As existing benchmarks for text-to-3D indoor scene generation remain limited in scale and diversity, we collect a new large-scale diverse dataset that contains 65 scene types and 3,250 instructions with diverse sizes, layouts, and styles, named 3DTindo-bench, to better assess the capability of the state-of-the-art models. Our experiments show that our method generates more realistic 3D scenes than state-of-the-art approaches.

DongQing Liu, MengShi Qi, HongWei Ji

CoVR-R studies reason-aware composed video retrieval: given a reference video and an edit instruction, the system must retrieve the target video that satisfies the edit. The main difficulty is that the target is not described directly; it must be inferred from fine-grained changes in object identity, action order, final state, hand interaction, and scene transition. We build a zero-shot reason-then-retrieve pipeline around Qwen3.5-27B. For each gallery video, the model generates a retrieval-oriented structured description and a dense embedding by pooling generated-token hidden states with token-dependent weights. For each query, the model first performs edit reasoning over the reference video and instruction, then generates a target-video description whose hidden states serve as the query embedding. We complement dense retrieval with a TF-IDF branch over the generated texts and fuse the two rankings with split-specific weights. On validation, the current best submission reaches 80.81 at R@1, 94.86 at R@5, 97.11 at R@10, and 98.59 at R@50. On the blind test split, it reaches 89.73 at R@1, 95.79 at R@5, 96.63 at R@10, and 97.98 at R@50.

Yilin Ou, Mengshi Qi, Huadong Ma

The VRR-QA challenge evaluates visual relational reasoning in videos, where answers often depend on implicit spatial relations, event boundaries, target identity, and dialogue context rather than a single salient frame. We present a test-time reasoning pipeline built around a strong GPT-5.5 video QA solver and a set of question-aware evidence ledgers. The initial solver answers each question from a uniform video representation, while routed ledgers are prompted to make the required targets, count units, reference frames, and temporal or spatial scope explicit for counting, spatial, endpoint, viewpoint, and dialogue reasoning. External tools such as open-vocabulary detection, depth cues, pair crops, ASR, and scene-graph ledgers are used only as evidence sources. A conservative gate keeps the current answer unless independent evidence uniquely supports a different option. The final evidence-gated pipeline achieves 92.95% overall accuracy and 93.79% macro accuracy on the challenge test split.

Rui Wang, Mengshi Qi, Yingxia Shao et al.

Monitoring blood pressure with non-invasive sensors has gained popularity for providing comfortable user experiences, one of which is a significant function of smart wearables. Although providing a comfortable user experience, such methods are suffering from the demand for a significant amount of realistic data to train an individual model for each subject, especially considering the invasive or obtrusive BP ground-truth measurements. To tackle this challenge, we introduce a novel physics-informed temporal network~(PITN) with adversarial contrastive learning to enable precise BP estimation with very limited data. Specifically, we first enhance the physics-informed neural network~(PINN) with the temporal block for investigating BP dynamics' multi-periodicity for personal cardiovascular cycle modeling and temporal variation. We then employ adversarial training to generate extra physiological time series data, improving PITN's robustness in the face of sparse subject-specific training data. Furthermore, we utilize contrastive learning to capture the discriminative variations of cardiovascular physiologic phenomena. This approach aggregates physiological signals with similar blood pressure values in latent space while separating clusters of samples with dissimilar blood pressure values. Experiments on three widely-adopted datasets with different modailties (\emph{i.e.,} bioimpedance, PPG, millimeter-wave) demonstrate the superiority and effectiveness of the proposed methods over previous state-of-the-art approaches. The code is available at~\url{https://github.com/Zest86/ACL-PITN}.

Changsheng Lv, Shuai Zhang, Yapeng Tian et al.

In this paper, we propose a Disentangled Counterfactual Learning~(DCL) approach for physical audiovisual commonsense reasoning. The task aims to infer objects' physics commonsense based on both video and audio input, with the main challenge is how to imitate the reasoning ability of humans. Most of the current methods fail to take full advantage of different characteristics in multi-modal data, and lacking causal reasoning ability in models impedes the progress of implicit physical knowledge inferring. To address these issues, our proposed DCL method decouples videos into static (time-invariant) and dynamic (time-varying) factors in the latent space by the disentangled sequential encoder, which adopts a variational autoencoder (VAE) to maximize the mutual information with a contrastive loss function. Furthermore, we introduce a counterfactual learning module to augment the model's reasoning ability by modeling physical knowledge relationships among different objects under counterfactual intervention. Our proposed method is a plug-and-play module that can be incorporated into any baseline. In experiments, we show that our proposed method improves baseline methods and achieves state-of-the-art performance. Our source code is available at https://github.com/Andy20178/DCL.

Haowen Wang, Mingyuan Wang, Zhengping Che et al.

The raw depth image captured by the indoor depth sensor usually has an extensive range of missing depth values due to inherent limitations such as the inability to perceive transparent objects and limited distance range. The incomplete depth map burdens many downstream vision tasks, and a rising number of depth completion methods have been proposed to alleviate this issue. While most existing methods can generate accurate dense depth maps from sparse and uniformly sampled depth maps, they are not suitable for complementing the large contiguous regions of missing depth values, which is common and critical. In this paper, we design a novel two-branch end-to-end fusion network, which takes a pair of RGB and incomplete depth images as input to predict a dense and completed depth map. The first branch employs an encoder-decoder structure to regress the local dense depth values from the raw depth map, with the help of local guidance information extracted from the RGB image. In the other branch, we propose an RGB-depth fusion GAN to transfer the RGB image to the fine-grained textured depth map. We adopt adaptive fusion modules named W-AdaIN to propagate the features across the two branches, and we append a confidence fusion head to fuse the two outputs of the branches for the final depth map. Extensive experiments on NYU-Depth V2 and SUN RGB-D demonstrate that our proposed method clearly improves the depth completion performance, especially in a more realistic setting of indoor environments with the help of the pseudo depth map.

Pengfei Zhu, Mengshi Qi, Xia Li et al.

Predicting attention regions of interest is an important yet challenging task for self-driving systems. Existing methodologies rely on large-scale labeled traffic datasets that are labor-intensive to obtain. Besides, the huge domain gap between natural scenes and traffic scenes in current datasets also limits the potential for model training. To address these challenges, we are the first to introduce an unsupervised way to predict self-driving attention by uncertainty modeling and driving knowledge integration. Our approach's Uncertainty Mining Branch (UMB) discovers commonalities and differences from multiple generated pseudo-labels achieved from models pre-trained on natural scenes by actively measuring the uncertainty. Meanwhile, our Knowledge Embedding Block (KEB) bridges the domain gap by incorporating driving knowledge to adaptively refine the generated pseudo-labels. Quantitative and qualitative results with equivalent or even more impressive performance compared to fully-supervised state-of-the-art approaches across all three public datasets demonstrate the effectiveness of the proposed method and the potential of this direction. The code will be made publicly available.

Zhe Zhao, Mengshi Qi, Huadong Ma

Generating realistic human grasps is a crucial yet challenging task for applications involving object manipulation in computer graphics and robotics. Existing methods often struggle with generating fine-grained realistic human grasps that ensure all fingers effectively interact with objects, as they focus on encoding hand with the whole representation and then estimating both hand posture and position in a single step. In this paper, we propose a novel Decomposed Vector-Quantized Variational Autoencoder (DVQ-VAE) to address this limitation by decomposing hand into several distinct parts and encoding them separately. This part-aware decomposed architecture facilitates more precise management of the interaction between each component of hand and object, enhancing the overall reality of generated human grasps. Furthermore, we design a newly dual-stage decoding strategy, by first determining the type of grasping under skeletal physical constraints, and then identifying the location of the grasp, which can greatly improve the verisimilitude as well as adaptability of the model to unseen hand-object interaction. In experiments, our model achieved about 14.1% relative improvement in the quality index compared to the state-of-the-art methods in four widely-adopted benchmarks. Our source code is available at https://github.com/florasion/D-VQVAE.

Haowen Wang, Zhengping Che, Yufan Yang et al.

Raw depth images captured in indoor scenarios frequently exhibit extensive missing values due to the inherent limitations of the sensors and environments. For example, transparent materials frequently elude detection by depth sensors; surfaces may introduce measurement inaccuracies due to their polished textures, extended distances, and oblique incidence angles from the sensor. The presence of incomplete depth maps imposes significant challenges for subsequent vision applications, prompting the development of numerous depth completion techniques to mitigate this problem. Numerous methods excel at reconstructing dense depth maps from sparse samples, but they often falter when faced with extensive contiguous regions of missing depth values, a prevalent and critical challenge in indoor environments. To overcome these challenges, we design a novel two-branch end-to-end fusion network named RDFC-GAN, which takes a pair of RGB and incomplete depth images as input to predict a dense and completed depth map. The first branch employs an encoder-decoder structure, by adhering to the Manhattan world assumption and utilizing normal maps from RGB-D information as guidance, to regress the local dense depth values from the raw depth map. The other branch applies an RGB-depth fusion CycleGAN, adept at translating RGB imagery into detailed, textured depth maps while ensuring high fidelity through cycle consistency. We fuse the two branches via adaptive fusion modules named W-AdaIN and train the model with the help of pseudo depth maps. Comprehensive evaluations on NYU-Depth V2 and SUN RGB-D datasets show that our method significantly enhances depth completion performance particularly in realistic indoor settings.

Wulian Yun, Mengshi Qi, Chuanming Wang et al.

Video denoising aims to recover high-quality frames from the noisy video. While most existing approaches adopt convolutional neural networks~(CNNs) to separate the noise from the original visual content, however, CNNs focus on local information and ignore the interactions between long-range regions in the frame. Furthermore, most related works directly take the output after basic spatio-temporal denoising as the final result, leading to neglect the fine-grained denoising process. In this paper, we propose a Dual-stage Spatial-Channel Transformer for coarse-to-fine video denoising, which inherits the advantages of both Transformer and CNNs. Specifically, DSCT is proposed based on a progressive dual-stage architecture, namely a coarse-level and a fine-level stage to extract dynamic features and static features, respectively. At both stages, a Spatial-Channel Encoding Module is designed to model the long-range contextual dependencies at both spatial and channel levels. Meanwhile, we design a Multi-Scale Residual Structure to preserve multiple aspects of information at different stages, which contains a Temporal Features Aggregation Module to summarize the dynamic representation. Extensive experiments on four publicly available datasets demonstrate our proposed method achieves significant improvements compared to the state-of-the-art methods.

Wulian Yun, Mengshi Qi, Chuanming Wang et al.

Weakly-supervised temporal action localization aims to locate action regions and identify action categories in untrimmed videos simultaneously by taking only video-level labels as the supervision. Pseudo label generation is a promising strategy to solve the challenging problem, but the current methods ignore the natural temporal structure of the video that can provide rich information to assist such a generation process. In this paper, we propose a novel weakly-supervised temporal action localization method by inferring salient snippet-feature. First, we design a saliency inference module that exploits the variation relationship between temporal neighbor snippets to discover salient snippet-features, which can reflect the significant dynamic change in the video. Secondly, we introduce a boundary refinement module that enhances salient snippet-features through the information interaction unit. Then, a discrimination enhancement module is introduced to enhance the discriminative nature of snippet-features. Finally, we adopt the refined snippet-features to produce high-fidelity pseudo labels, which could be used to supervise the training of the action localization network. Extensive experiments on two publicly available datasets, i.e., THUMOS14 and ActivityNet v1.3, demonstrate our proposed method achieves significant improvements compared to the state-of-the-art methods.

Mengshi Qi, Yeteng Wu, Xianlin Zhang et al.

Evaluating whether human action is standard or not and providing reasonable feedback to improve action standardization is very crucial but challenging in real-world scenarios. However, current video understanding methods are mainly concerned with what and where the action is, which is unable to meet the requirements. Meanwhile, most of the existing datasets lack the labels indicating the degree of action standardization, and the action quality assessment datasets lack explainability and detailed feedback. Therefore, we define a new Human Action Form Assessment (AFA) task, and introduce a new diverse dataset CoT-AFA, which contains a large scale of fitness and martial arts videos with multi-level annotations for comprehensive video analysis. We enrich the CoT-AFA dataset with a novel Chain-of-Thought explanation paradigm. Instead of offering isolated feedback, our explanations provide a complete reasoning process--from identifying an action step to analyzing its outcome and proposing a concrete solution. Furthermore, we propose a framework named Explainable Fitness Assessor, which can not only judge an action but also explain why and provide a solution. This framework employs two parallel processing streams and a dynamic gating mechanism to fuse visual and semantic information, thereby boosting its analytical capabilities. The experimental results demonstrate that our method has achieved improvements in explanation generation (e.g., +16.0% in CIDEr), action classification (+2.7% in accuracy) and quality assessment (+2.1% in accuracy), revealing great potential of CoT-AFA for future studies. Our dataset and source code is available at https://github.com/MICLAB-BUPT/EFA.

Wulian Yun, Mengshi Qi, Fei Peng et al.

Existing action quality assessment (AQA) methods often require a large number of label annotations for fully supervised learning, which are laborious and expensive. In practice, the labeled data are difficult to obtain because the AQA annotation process requires domain-specific expertise. In this paper, we propose a novel semi-supervised method, which can be utilized for better assessment of the AQA task by exploiting a large amount of unlabeled data and a small portion of labeled data. Differing from the traditional teacher-student network, we propose a teacher-reference-student architecture to learn both unlabeled and labeled data, where the teacher network and the reference network are used to generate pseudo-labels for unlabeled data to supervise the student network. Specifically, the teacher predicts pseudo-labels by capturing high-level features of unlabeled data. The reference network provides adequate supervision of the student network by referring to additional action information. Moreover, we introduce confidence memory to improve the reliability of pseudo-labels by storing the most accurate ever output of the teacher network and reference network. To validate our method, we conduct extensive experiments on three AQA benchmark datasets. Experimental results show that our method achieves significant improvements and outperforms existing semi-supervised AQA methods.

Wei Deng, Mengshi Qi, Huadong Ma

Large Vision-Language Models (VLMs), such as GPT-4, have achieved remarkable success across various fields. However, there are few studies on 3D indoor scene generation with VLMs. This paper considers this task as a planning problem subject to spatial and layout common sense constraints. To solve the problem with a VLM, we propose a new global-local tree search algorithm. Globally, the method places each object sequentially and explores multiple placements during each placement process, where the problem space is represented as a tree. To reduce the depth of the tree, we decompose the scene structure hierarchically, i.e. room level, region level, floor object level, and supported object level. The algorithm independently generates the floor objects in different regions and supported objects placed on different floor objects. Locally, we also decompose the sub-task, the placement of each object, into multiple steps. The algorithm searches the tree of problem space. To leverage the VLM model to produce positions of objects, we discretize the top-down view space as a dense grid and fill each cell with diverse emojis to make to cells distinct. We prompt the VLM with the emoji grid and the VLM produces a reasonable location for the object by describing the position with the name of emojis. The quantitative and qualitative experimental results illustrate our approach generates more plausible 3D scenes than state-of-the-art approaches. Our source code is available at https://github.com/dw-dengwei/TreeSearchGen .

Dacheng Liao, Mengshi Qi, Peng Shu et al.

Current Vision-Language Model (VLM)-based end-to-end autonomous driving systems often leverage large language models to generate driving decisions directly based on their understanding of the current scene. However, such systems introduce multiple risks in real-world driving scenarios. To evaluate whether VLMs are truly viable for autonomous driving, we introduce RoboDriveBench, the first robustness benchmark focused on end-to-end trajectory prediction tasks. This benchmark systematically evaluates two critical categories of real-world challenges for VLM-based end-to-end autonomous driving systems through 11 simulated scenarios encompassing various corruption types, including 6 scenarios of sensor corruption caused by environmental variations, along with 5 cases of prompt corruption resulting from human intervention and data transmission failures. Each corruption type includes 250 unique driving scenarios and 5,689 frames, resulting in 64,559 total trajectory prediction cases per evaluation. To overcome these real-world challenges, we propose a novel VLM-based autonomous driving framework called RoboDriveVLM, which enhances robustness by mapping more multimodal data-e.g., lidar and radar-into a unified latent space. Furthermore, we introduce a new Test-Time Adaptation (TTA) method based on cross-modal knowledge distillation to improve the robustness of VLM-based autonomous driving systems. Through extensive experiments, our work highlights the limitations of current VLM-based end-to-end autonomous driving systems and provides a more reliable solution for real-world deployment. Source code and datasets will be released.

Huiyuan Fu, Kuilong Cui, Chuanming Wang et al.

With the rapid advancements in deep learning technologies, person re-identification (ReID) has witnessed remarkable performance improvements. However, the majority of prior works have traditionally focused on solving the problem via extracting features solely from a single perspective, such as uniform partitioning, hard attention mechanisms, or semantic masks. While these approaches have demonstrated efficacy within specific contexts, they fall short in diverse situations. In this paper, we propose a novel approach, Mutual Distillation Learning For Person Re-identification (termed as MDPR), which addresses the challenging problem from multiple perspectives within a single unified model, leveraging the power of mutual distillation to enhance the feature representations collectively. Specifically, our approach encompasses two branches: a hard content branch to extract local features via a uniform horizontal partitioning strategy and a Soft Content Branch to dynamically distinguish between foreground and background and facilitate the extraction of multi-granularity features via a carefully designed attention mechanism. To facilitate knowledge exchange between these two branches, a mutual distillation and fusion process is employed, promoting the capability of the outputs of each branch. Extensive experiments are conducted on widely used person ReID datasets to validate the effectiveness and superiority of our approach. Notably, our method achieves an impressive $88.7\%/94.4\%$ in mAP/Rank-1 on the DukeMTMC-reID dataset, surpassing the current state-of-the-art results. Our source code is available at https://github.com/KuilongCui/MDPR.

Mengshi Qi, Hao Ye, Jiaxuan Peng et al.

Action Quality Assessment (AQA), which aims at automatic and fair evaluation of athletic performance, has gained increasing attention in recent years. However, athletes are often in rapid movement and the corresponding visual appearance variances are subtle, making it challenging to capture fine-grained pose differences and leading to poor estimation performance. Furthermore, most common AQA tasks, such as diving in sports, are usually divided into multiple sub-actions, each of which contains different durations. However, existing methods focus on segmenting the video into fixed frames, which disrupts the temporal continuity of sub-actions resulting in unavoidable prediction errors. To address these challenges, we propose a novel action quality assessment method through hierarchically pose-guided multi-stage contrastive regression. Firstly, we introduce a multi-scale dynamic visual-skeleton encoder to capture fine-grained spatio-temporal visual and skeletal features. Then, a procedure segmentation network is introduced to separate different sub-actions and obtain segmented features. Afterwards, the segmented visual and skeletal features are both fed into a multi-modal fusion module as physics structural priors, to guide the model in learning refined activity similarities and variances. Finally, a multi-stage contrastive learning regression approach is employed to learn discriminative representations and output prediction results. In addition, we introduce a newly-annotated FineDiving-Pose Dataset to improve the current low-quality human pose labels. In experiments, the results on FineDiving and MTL-AQA datasets demonstrate the effectiveness and superiority of our proposed approach. Our source code and dataset are available at https://github.com/Lumos0507/HP-MCoRe.

Mengshi Qi, Yuxin Yang, Huadong Ma

Effective modeling of group interactions and dynamic semantic intentions is crucial for forecasting behaviors like trajectories or movements. In complex scenarios like sports, agents' trajectories are influenced by group interactions and intentions, including team strategies and opponent actions. To this end, we propose a novel diffusion-based trajectory prediction framework that integrates group-level interactions into a conditional diffusion model, enabling the generation of diverse trajectories aligned with specific group activity. To capture dynamic semantic intentions, we frame group interaction prediction as a cooperative game, using Banzhaf interaction to model cooperation trends. We then fuse semantic intentions with enhanced agent embeddings, which are refined through both global and local aggregation. Furthermore, we expand the NBA SportVU dataset by adding human annotations of team-level tactics for trajectory and tactic prediction tasks. Extensive experiments on three widely-adopted datasets demonstrate that our model outperforms state-of-the-art methods. Our source code and data are available at https://github.com/aurora-xin/Group2Int-trajectory.

Hao Ye, Mengshi Qi, Zhaohong Liu et al.

In this work, we study how vision-language models (VLMs) can be utilized to enhance the safety for the autonomous driving system, including perception, situational understanding, and path planning. However, existing research has largely overlooked the evaluation of these models in traffic safety-critical driving scenarios. To bridge this gap, we create the benchmark (SafeDrive228K) and propose a new baseline based on VLM with knowledge graph-based retrieval-augmented generation (SafeDriveRAG) for visual question answering (VQA). Specifically, we introduce SafeDrive228K, the first large-scale multimodal question-answering benchmark comprising 228K examples across 18 sub-tasks. This benchmark encompasses a diverse range of traffic safety queries, from traffic accidents and corner cases to common safety knowledge, enabling a thorough assessment of the comprehension and reasoning abilities of the models. Furthermore, we propose a plug-and-play multimodal knowledge graph-based retrieval-augmented generation approach that employs a novel multi-scale subgraph retrieval algorithm for efficient information retrieval. By incorporating traffic safety guidelines collected from the Internet, this framework further enhances the model's capacity to handle safety-critical situations. Finally, we conduct comprehensive evaluations on five mainstream VLMs to assess their reliability in safety-sensitive driving tasks. Experimental results demonstrate that integrating RAG significantly improves performance, achieving a +4.73% gain in Traffic Accidents tasks, +8.79% in Corner Cases tasks and +14.57% in Traffic Safety Commonsense across five mainstream VLMs, underscoring the potential of our proposed benchmark and methodology for advancing research in traffic safety. Our source code and data are available at https://github.com/Lumos0507/SafeDriveRAG.

Hongwei Ji, Wulian Yun, Mengshi Qi et al.

Traditional temporal action localization (TAL) methods rely on large amounts of detailed annotated data, whereas few-shot TAL reduces this dependence by using only a few training samples to identify unseen action categories. However, existing few-shot TAL methods typically focus solely on video-level information, neglecting textual information, which can provide valuable semantic support for the localization task. Therefore, we propose a new few-shot temporal action localization method by Chain-of-Thought textual reasoning to improve localization performance. Specifically, we design a novel few-shot learning framework that leverages textual semantic information to enhance the model's ability to capture action commonalities and variations, which includes a semantic-aware text-visual alignment module designed to align the query and support videos at different levels. Meanwhile, to better express the temporal dependencies and causal relationships between actions at the textual level to assist action localization, we design a Chain of Thought (CoT)-like reasoning method that progressively guides the Vision Language Model (VLM) and Large Language Model (LLM) to generate CoT-like text descriptions for videos. The generated texts can capture more variance of action than visual features. We conduct extensive experiments on the publicly available ActivityNet1.3 and THUMOS14 datasets. We introduce the first dataset named Human-related Anomaly Localization and explore the application of the TAL task in human anomaly detection. The experimental results demonstrate that our proposed method significantly outperforms existing methods in single-instance and multi-instance scenarios. We will release our code, data and benchmark.

Mengshi Qi, Pengfei Zhu, Xiangtai Li et al.

Given a single labeled example, in-context segmentation aims to segment corresponding objects. This setting, known as one-shot segmentation in few-shot learning, explores the segmentation model's generalization ability and has been applied to various vision tasks, including scene understanding and image/video editing. While recent Segment Anything Models have achieved state-of-the-art results in interactive segmentation, these approaches are not directly applicable to in-context segmentation. In this work, we propose the Dual Consistency SAM (DC-SAM) method based on prompt-tuning to adapt SAM and SAM2 for in-context segmentation of both images and videos. Our key insights are to enhance the features of the SAM's prompt encoder in segmentation by providing high-quality visual prompts. When generating a mask prior, we fuse the SAM features to better align the prompt encoder. Then, we design a cycle-consistent cross-attention on fused features and initial visual prompts. Next, a dual-branch design is provided by using the discriminative positive and negative prompts in the prompt encoder. Furthermore, we design a simple mask-tube training strategy to adopt our proposed dual consistency method into the mask tube. Although the proposed DC-SAM is primarily designed for images, it can be seamlessly extended to the video domain with the support of SAM2. Given the absence of in-context segmentation in the video domain, we manually curate and construct the first benchmark from existing video segmentation datasets, named In-Context Video Object Segmentation (IC-VOS), to better assess the in-context capability of the model. Extensive experiments demonstrate that our method achieves 55.5 (+1.4) mIoU on COCO-20i, 73.0 (+1.1) mIoU on PASCAL-5i, and a J&F score of 71.52 on the proposed IC-VOS benchmark. Our source code and benchmark are available at https://github.com/zaplm/DC-SAM.

Pengfei Zhu, Peng Shu, Mengshi Qi et al.

Accurate prediction of future trajectories of traffic agents is essential for ensuring safe autonomous driving. However, partially observed trajectories can significantly degrade the performance of even state-of-the-art models. Previous approaches often rely on knowledge distillation to transfer features from fully observed trajectories to partially observed ones. This involves firstly training a fully observed model and then using a distillation process to create the final model. While effective, they require multi-stage training, making the training process very expensive. Moreover, knowledge distillation can lead to a performance degradation of the model. In this paper, we introduce a Target-driven Self-Distillation method (TSD) for motion forecasting. Our method leverages predicted accurate targets to guide the model in making predictions under partial observation conditions. By employing self-distillation, the model learns from the feature distributions of both fully observed and partially observed trajectories during a single end-to-end training process. This enhances the model's ability to predict motion accurately in both fully observed and partially observed scenarios. We evaluate our method on multiple datasets and state-of-the-art motion forecasting models. Extensive experimental results demonstrate that our approach achieves significant performance improvements in both settings. To facilitate further research, we will release our code and model checkpoints.

Mengshi Qi, Xiaoyang Bi, Pengfei Zhu et al.

Robustly predicting attention regions of interest for self-driving systems is crucial for driving safety but presents significant challenges due to the labor-intensive nature of obtaining large-scale attention labels and the domain gap between self-driving scenarios and natural scenes. These challenges are further exacerbated by complex traffic environments, including camera corruption under adverse weather, noise interferences, and central bias from long-tail distributions. To address these issues, we propose a robust unsupervised attention prediction method. An Uncertainty Mining Branch refines predictions by analyzing commonalities and differences across multiple pre-trained models on natural scenes, while a Knowledge Embedding Block bridges the domain gap by incorporating driving knowledge to adaptively enhance pseudo-labels. Additionally, we introduce RoboMixup, a novel data augmentation method that improves robustness against corruption through soft attention and dynamic augmentation, and mitigates central bias by integrating random cropping into Mixup as a regularizer. To systematically evaluate robustness in self-driving attention prediction, we introduce the DriverAttention-C benchmark, comprising over 100k frames across three subsets: BDD-A-C, DR(eye)VE-C, and DADA-2000-C. Our method achieves performance equivalent to or surpassing fully supervised state-of-the-art approaches on three public datasets and the proposed robustness benchmark, reducing relative corruption degradation by 58.8% and 52.8%, and improving central bias robustness by 12.4% and 11.4% in KLD and CC metrics, respectively. Code and data are available at https://github.com/zaplm/DriverAttention.

Jiaxuan Peng, Mengshi Qi, Dong Zhao et al.

3D human pose estimation (3D HPE) has emerged as a prominent research topic, particularly in the realm of RGB-based methods. However, RGB images are susceptible to limitations such as sensitivity to lighting conditions and potential user discomfort. Consequently, multi-modal sensing, which leverages non-intrusive sensors, is gaining increasing attention. Nevertheless, multi-modal 3D HPE still faces challenges, including modality imbalance and the imperative for continual learning. In this work, we introduce a novel balanced continual multi-modal learning method for 3D HPE, which harnesses the power of RGB, LiDAR, mmWave, and WiFi. Specifically, we propose a Shapley value-based contribution algorithm to quantify the contribution of each modality and identify modality imbalance. To address this imbalance, we employ a re-learning strategy. Furthermore, recognizing that raw data is prone to noise contamination, we develop a novel denoising continual learning approach. This approach incorporates a noise identification and separation module to mitigate the adverse effects of noise and collaborates with the balanced learning strategy to enhance optimization. Additionally, an adaptive EWC mechanism is employed to alleviate catastrophic forgetting. We conduct extensive experiments on the widely-adopted multi-modal dataset, MM-Fi, which demonstrate the superiority of our approach in boosting 3D pose estimation and mitigating catastrophic forgetting in complex scenarios. We will release our codes.

Dacheng Liao, Mengshi Qi, Liang Liu et al.

In current open real-world autonomous driving scenarios, challenges such as sensor failure and extreme weather conditions hinder the generalization of most autonomous driving perception models to these unseen domain due to the domain shifts between the test and training data. As the parameter scale of autonomous driving perception models grows, traditional test-time adaptation (TTA) methods become unstable and often degrade model performance in most scenarios. To address these challenges, this paper proposes two new robust methods to improve the Batch Normalization with TTA for object detection in autonomous driving: (1) We introduce a LearnableBN layer based on Generalized-search Entropy Minimization (GSEM) method. Specifically, we modify the traditional BN layer by incorporating auxiliary learnable parameters, which enables the BN layer to dynamically update the statistics according to the different input data. (2) We propose a new semantic-consistency based dual-stage-adaptation strategy, which encourages the model to iteratively search for the optimal solution and eliminates unstable samples during the adaptation process. Extensive experiments on the NuScenes-C dataset shows that our method achieves a maximum improvement of about 8% using BEVFormer as the baseline model across six corruption types and three levels of severity. We will make our source code available soon.

Changsheng Lv, Mengshi Qi, Liang Liu et al.

Understanding the traffic scenes and then generating high-definition (HD) maps present significant challenges in autonomous driving. In this paper, we defined a novel Traffic Topology Scene Graph, a unified scene graph explicitly modeling the lane, controlled and guided by different road signals (e.g., right turn), and topology relationships among them, which is always ignored by previous high-definition (HD) mapping methods. For the generation of T2SG, we propose TopoFormer, a novel one-stage Topology Scene Graph TransFormer with two newly designed layers. Specifically, TopoFormer incorporates a Lane Aggregation Layer (LAL) that leverages the geometric distance among the centerline of lanes to guide the aggregation of global information. Furthermore, we proposed a Counterfactual Intervention Layer (CIL) to model the reasonable road structure ( e.g., intersection, straight) among lanes under counterfactual intervention. Then the generated T2SG can provide a more accurate and explainable description of the topological structure in traffic scenes. Experimental results demonstrate that TopoFormer outperforms existing methods on the T2SG generation task, and the generated T2SG significantly enhances traffic topology reasoning in downstream tasks, achieving a state-of-the-art performance of 46.3 OLS on the OpenLane-V2 benchmark. We will release our source code and model.

Qi An, Mengshi Qi, Huadong Ma

In recent years, there has been growing interest in the video-based action quality assessment (AQA). Most existing methods typically solve AQA problem by considering the entire video yet overlooking the inherent stage-level characteristics of actions. To address this issue, we design a novel Multi-stage Contrastive Regression (MCoRe) framework for the AQA task. This approach allows us to efficiently extract spatial-temporal information, while simultaneously reducing computational costs by segmenting the input video into multiple stages or procedures. Inspired by the graph contrastive learning, we propose a new stage-wise contrastive learning loss function to enhance performance. As a result, MCoRe demonstrates the state-of-the-art result so far on the widely-adopted fine-grained AQA dataset.

Yuxin Yang, Pengfei Zhu, Mengshi Qi et al.

Human trajectory forecasting is a critical challenge in fields such as robotics and autonomous driving. Due to the inherent uncertainty of human actions and intentions in real-world scenarios, various unexpected occurrences may arise. To uncover latent motion patterns in human behavior, we introduce a novel memory-based method, named Motion Pattern Priors Memory Network. Our method involves constructing a memory bank derived from clustered prior knowledge of motion patterns observed in the training set trajectories. We introduce an addressing mechanism to retrieve the matched pattern and the potential target distributions for each prediction from the memory bank, which enables the identification and retrieval of natural motion patterns exhibited by agents, subsequently using the target priors memory token to guide the diffusion model to generate predictions. Extensive experiments validate the effectiveness of our approach, achieving state-of-the-art trajectory prediction accuracy. The code will be made publicly available.

Yuxin Lin, Mengshi Qi, Liang Liu et al.

In this paper, we propose a novel approach for solving the Visual Question Answering (VQA) task in autonomous driving by integrating Vision-Language Models (VLMs) with continual learning. In autonomous driving, VQA plays a vital role in enabling the system to understand and reason about its surroundings. However, traditional models often struggle with catastrophic forgetting when sequentially exposed to new driving tasks, such as perception, prediction, and planning, each requiring different forms of knowledge. To address this challenge, we present a novel continual learning framework that combines VLMs with selective memory replay and knowledge distillation, reinforced by task-specific projection layer regularization. The knowledge distillation allows a previously trained model to act as a "teacher" to guide the model through subsequent tasks, minimizing forgetting. Meanwhile, task-specific projection layers calculate the loss based on the divergence of feature representations, ensuring continuity in learning and reducing the shift between tasks. Evaluated on the DriveLM dataset, our framework shows substantial performance improvements, with gains ranging from 21.40% to 32.28% across various metrics. These results highlight the effectiveness of combining continual learning with VLMs in enhancing the resilience and reliability of VQA systems in autonomous driving. We will release our source code.

Mengshi Qi, Changsheng Lv, Huadong Ma

In this paper, we propose a new Robust Disentangled Counterfactual Learning (RDCL) approach for physical audiovisual commonsense reasoning. The task aims to infer objects' physics commonsense based on both video and audio input, with the main challenge being how to imitate the reasoning ability of humans, even under the scenario of missing modalities. Most of the current methods fail to take full advantage of different characteristics in multi-modal data, and lacking causal reasoning ability in models impedes the progress of implicit physical knowledge inferring. To address these issues, our proposed RDCL method decouples videos into static (time-invariant) and dynamic (time-varying) factors in the latent space by the disentangled sequential encoder, which adopts a variational autoencoder (VAE) to maximize the mutual information with a contrastive loss function. Furthermore, we introduce a counterfactual learning module to augment the model's reasoning ability by modeling physical knowledge relationships among different objects under counterfactual intervention. To alleviate the incomplete modality data issue, we introduce a robust multimodal learning method to recover the missing data by decomposing the shared features and model-specific features. Our proposed method is a plug-and-play module that can be incorporated into any baseline including VLMs. In experiments, we show that our proposed method improves the reasoning accuracy and robustness of baseline methods and achieves the state-of-the-art performance.

Chuanming Wang, Yuxin Yang, Mengshi Qi et al.

Object re-identification (ReID) is committed to searching for objects of the same identity across cameras, and its real-world deployment is gradually increasing. Current ReID methods assume that the deployed system follows the centralized processing paradigm, i.e., all computations are conducted in the cloud server and edge devices are only used to capture images. As the number of videos experiences a rapid escalation, this paradigm has become impractical due to the finite computational resources in the cloud server. Therefore, the ReID system should be converted to fit in the cloud-edge collaborative processing paradigm, which is crucial to boost its scalability and practicality. However, current works lack relevant research on this important specific issue, making it difficult to adapt them into a cloud-edge framework effectively. In this paper, we propose a cloud-edge collaborative inference framework for ReID systems, aiming to expedite the return of the desired image captured by the camera to the cloud server by learning the spatial-temporal correlations among objects. In the system, a Distribution-aware Correlation Modeling network (DaCM) is particularly proposed to embed the spatial-temporal correlations of the camera network implicitly into a graph structure, and it can be applied 1) in the cloud to regulate the size of the upload window and 2) on the edge device to adjust the sequence of images, respectively. Notably, the proposed DaCM can be seamlessly combined with traditional ReID methods, enabling their application within our proposed edge-cloud collaborative framework. Extensive experiments demonstrate that our method obviously reduces transmission overhead and significantly improves performance.

Zijian Fu, Changsheng Lv, Mengshi Qi et al.

In this paper, we propose a novel Multi-Modal Scene Graph with Kolmogorov-Arnold Expert Network for Audio-Visual Question Answering (SHRIKE). The task aims to mimic human reasoning by extracting and fusing information from audio-visual scenes, with the main challenge being the identification of question-relevant cues from the complex audio-visual content. Existing methods fail to capture the structural information within video, and suffer from insufficient fine-grained modeling of multi-modal features. To address these issues, we are the first to introduce a new multi-modal scene graph that explicitly models the objects and their relationship as a visually grounded, structured representation of the audio-visual scene. Furthermore, we design a Kolmogorov-Arnold Network~(KAN)-based Mixture of Experts (MoE) to enhance the expressive power of the temporal integration stage. This enables more fine-grained modeling of cross-modal interactions within the question-aware fused audio-visual representation, leading to capture richer and more nuanced patterns and then improve temporal reasoning performance. We evaluate the model on the established MUSIC-AVQA and MUSIC-AVQA v2 benchmarks, where it achieves state-of-the-art performance. Code and model checkpoints will be publicly released.

Silvio Giancola, Anthony Cioppa, Marc Gutiérrez-Pérez et al.

The SoccerNet 2025 Challenges mark the fifth annual edition of the SoccerNet open benchmarking effort, dedicated to advancing computer vision research in football video understanding. This year's challenges span four vision-based tasks: (1) Team Ball Action Spotting, focused on detecting ball-related actions in football broadcasts and assigning actions to teams; (2) Monocular Depth Estimation, targeting the recovery of scene geometry from single-camera broadcast clips through relative depth estimation for each pixel; (3) Multi-View Foul Recognition, requiring the analysis of multiple synchronized camera views to classify fouls and their severity; and (4) Game State Reconstruction, aimed at localizing and identifying all players from a broadcast video to reconstruct the game state on a 2D top-view of the field. Across all tasks, participants were provided with large-scale annotated datasets, unified evaluation protocols, and strong baselines as starting points. This report presents the results of each challenge, highlights the top-performing solutions, and provides insights into the progress made by the community. The SoccerNet Challenges continue to serve as a driving force for reproducible, open research at the intersection of computer vision, artificial intelligence, and sports. Detailed information about the tasks, challenges, and leaderboards can be found at https://www.soccer-net.org, with baselines and development kits available at https://github.com/SoccerNet.

Changsheng Lv, Mengshi Qi, Zijian Fu et al.

In this paper, we introduce a novel method named Robo-SGG, i.e., Layout-Oriented Normalization and Restitution for Robust Scene Graph Generation. Compared to the existing SGG setting, the robust scene graph generation aims to perform inference on a diverse range of corrupted images, with the core challenge being the domain shift between the clean and corrupted images. Existing SGG methods suffer from degraded performance due to compromised visual features e.g., corruption interference or occlusions. To obtain robust visual features, we exploit the layout information, which is domain-invariant, to enhance the efficacy of existing SGG methods on corrupted images. Specifically, we employ Instance Normalization(IN) to filter out the domain-specific feature and recover the unchangeable structural features, i.e., the positional and semantic relationships among objects by the proposed Layout-Oriented Restitution. Additionally, we propose a Layout-Embedded Encoder (LEE) that augments the existing object and predicate encoders within the SGG framework, enriching the robust positional and semantic features of objects and predicates. Note that our proposed Robo-SGG module is designed as a plug-and-play component, which can be easily integrated into any baseline SGG model. Extensive experiments demonstrate that by integrating the state-of-the-art method into our proposed Robo-SGG, we achieve relative improvements of 5.6%, 8.0%, and 6.5% in mR@50 for PredCls, SGCls, and SGDet tasks on the VG-C dataset, respectively, and achieve new state-of-the-art performance in corruption scene graph generation benchmark (VG-C and GQA-C). We will release our source code and model.

Wulian Yun, Mengshi Qi, Fei Peng et al.

Conventional 2D human pose estimation methods typically require extensive labeled annotations, which are both labor-intensive and expensive. In contrast, semi-supervised 2D human pose estimation can alleviate the above problems by leveraging a large amount of unlabeled data along with a small portion of labeled data. Existing semi-supervised 2D human pose estimation methods update the network through backpropagation, ignoring crucial historical information from the previous training process. Therefore, we propose a novel semi-supervised 2D human pose estimation method by utilizing a newly designed Teacher-Reviewer-Student framework. Specifically, we first mimic the phenomenon that human beings constantly review previous knowledge for consolidation to design our framework, in which the teacher predicts results to guide the student's learning and the reviewer stores important historical parameters to provide additional supervision signals. Secondly, we introduce a Multi-level Feature Learning strategy, which utilizes the outputs from different stages of the backbone to estimate the heatmap to guide network training, enriching the supervisory information while effectively capturing keypoint relationships. Finally, we design a data augmentation strategy, i.e., Keypoint-Mix, to perturb pose information by mixing different keypoints, thus enhancing the network's ability to discern keypoints. Extensive experiments on publicly available datasets, demonstrate our method achieves significant improvements compared to the existing methods.

Mengshi Qi, Edoardo Remelli, Mathieu Salzmann et al.

Deep learning-solutions for hand-object 3D pose and shape estimation are now very effective when an annotated dataset is available to train them to handle the scenarios and lighting conditions they will encounter at test time. Unfortunately, this is not always the case, and one often has to resort to training them on synthetic data, which does not guarantee that they will work well in real situations. In this paper, we introduce an effective approach to addressing this challenge by exploiting 3D geometric constraints within a cycle generative adversarial network (CycleGAN) to perform domain adaptation. Furthermore, in contrast to most existing works, which fail to leverage the rich temporal information available in unlabeled real videos as a source of supervision, we propose to enforce short- and long-term temporal consistency to fine-tune the domain-adapted model in a self-supervised fashion. We will demonstrate that our approach outperforms state-of-the-art 3D hand-object joint reconstruction methods on three widely-used benchmarks and will make our code publicly available.

Mengshi Qi, Weijian Li, Zhengyuan Yang et al.

Scene graph generation refers to the task of automatically mapping an image into a semantic structural graph, which requires correctly labeling each extracted object and their interaction relationships. Despite the recent success in object detection using deep learning techniques, inferring complex contextual relationships and structured graph representations from visual data remains a challenging topic. In this study, we propose a novel Attentive Relational Network that consists of two key modules with an object detection backbone to approach this problem. The first module is a semantic transformation module utilized to capture semantic embedded relation features, by translating visual features and linguistic features into a common semantic space. The other module is a graph self-attention module introduced to embed a joint graph representation through assigning various importance weights to neighboring nodes. Finally, accurate scene graphs are produced by the relation inference module to recognize all entities and the corresponding relations. We evaluate our proposed method on the widely-adopted Visual Genome Dataset, and the results demonstrate the effectiveness and superiority of our model.