Jianbing Shen

Yucheng Zhou, Wei Tao, Yiwen Guo et al.

World models and multimodal large language models (MLLMs) provide complementary capabilities for predicting future outcomes from static visual observations. World models can generate concrete visual rollouts of possible futures, while MLLMs can reason abstractly over questions, goals, and rules. However, generated rollouts are stochastic and may be visually plausible but task-incorrect, making it necessary to determine when visual simulation is useful, whether a rollout is credible, and how it should influence the final answer. We formulate this problem as controlled concrete reasoning, where a model learns to invoke, verify, and integrate visual future simulation alongside abstract reasoning. To study this setting, we construct two human-verified benchmarks, VRQABench for controllable spatial lookahead and OpenWorldQA for open-domain physical prediction, and propose Privileged-Future On-Policy Self-Distillation (PF-OPSD). During training, PF-OPSD uses ground-truth future videos and answers only as teacher-side privileged context to evaluate on-policy concrete-reasoning trajectories, while the deployable student never observes true futures at test time. Experimental results show that PF-OPSD outperforms baseline by 10.6% and 10.9% on VRQABench and OpenWorldQA, respectively, while increasing robustness to noisy or conflicting rollouts. Our code and dataset are available at https://github.com/yczhou001/PF-OPSD.

Dongming Wu, Wencheng Han, Tiancai Wang et al.

Existing referring understanding tasks tend to involve the detection of a single text-referred object. In this paper, we propose a new and general referring understanding task, termed referring multi-object tracking (RMOT). Its core idea is to employ a language expression as a semantic cue to guide the prediction of multi-object tracking. To the best of our knowledge, it is the first work to achieve an arbitrary number of referent object predictions in videos. To push forward RMOT, we construct one benchmark with scalable expressions based on KITTI, named Refer-KITTI. Specifically, it provides 18 videos with 818 expressions, and each expression in a video is annotated with an average of 10.7 objects. Further, we develop a transformer-based architecture TransRMOT to tackle the new task in an online manner, which achieves impressive detection performance and outperforms other counterparts. The dataset and code will be available at https://github.com/wudongming97/RMOT.

Zhiyuan Liang, Tiancai Wang, Xiangyu Zhang et al.

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

Yan Wang, Junbo Yin, Wei Li et al.

LiDAR-based 3D object detection is an indispensable task in advanced autonomous driving systems. Though impressive detection results have been achieved by superior 3D detectors, they suffer from significant performance degeneration when facing unseen domains, such as different LiDAR configurations, different cities, and weather conditions. The mainstream approaches tend to solve these challenges by leveraging unsupervised domain adaptation (UDA) techniques. However, these UDA solutions just yield unsatisfactory 3D detection results when there is a severe domain shift, e.g., from Waymo (64-beam) to nuScenes (32-beam). To address this, we present a novel Semi-Supervised Domain Adaptation method for 3D object detection (SSDA3D), where only a few labeled target data is available, yet can significantly improve the adaptation performance. In particular, our SSDA3D includes an Inter-domain Adaptation stage and an Intra-domain Generalization stage. In the first stage, an Inter-domain Point-CutMix module is presented to efficiently align the point cloud distribution across domains. The Point-CutMix generates mixed samples of an intermediate domain, thus encouraging to learn domain-invariant knowledge. Then, in the second stage, we further enhance the model for better generalization on the unlabeled target set. This is achieved by exploring Intra-domain Point-MixUp in semi-supervised learning, which essentially regularizes the pseudo label distribution. Experiments from Waymo to nuScenes show that, with only 10% labeled target data, our SSDA3D can surpass the fully-supervised oracle model with 100% target label. Our code is available at https://github.com/yinjunbo/SSDA3D.

Runzhou Tao, Wencheng Han, Zhongying Qiu et al.

Monocular 3D object detection has become a mainstream approach in automatic driving for its easy application. A prominent advantage is that it does not need LiDAR point clouds during the inference. However, most current methods still rely on 3D point cloud data for labeling the ground truths used in the training phase. This inconsistency between the training and inference makes it hard to utilize the large-scale feedback data and increases the data collection expenses. To bridge this gap, we propose a new weakly supervised monocular 3D objection detection method, which can train the model with only 2D labels marked on images. To be specific, we explore three types of consistency in this task, i.e. the projection, multi-view and direction consistency, and design a weakly-supervised architecture based on these consistencies. Moreover, we propose a new 2D direction labeling method in this task to guide the model for accurate rotation direction prediction. Experiments show that our weakly-supervised method achieves comparable performance with some fully supervised methods. When used as a pre-training method, our model can significantly outperform the corresponding fully-supervised baseline with only 1/3 3D labels. https://github.com/weakmono3d/weakmono3d

Xiang Li, Junbo Yin, Botian Shi et al.

Image instance segmentation is a fundamental research topic in autonomous driving, which is crucial for scene understanding and road safety. Advanced learning-based approaches often rely on the costly 2D mask annotations for training. In this paper, we present a more artful framework, LiDAR-guided Weakly Supervised Instance Segmentation (LWSIS), which leverages the off-the-shelf 3D data, i.e., Point Cloud, together with the 3D boxes, as natural weak supervisions for training the 2D image instance segmentation models. Our LWSIS not only exploits the complementary information in multimodal data during training, but also significantly reduces the annotation cost of the dense 2D masks. In detail, LWSIS consists of two crucial modules, Point Label Assignment (PLA) and Graph-based Consistency Regularization (GCR). The former module aims to automatically assign the 3D point cloud as 2D point-wise labels, while the latter further refines the predictions by enforcing geometry and appearance consistency of the multimodal data. Moreover, we conduct a secondary instance segmentation annotation on the nuScenes, named nuInsSeg, to encourage further research on multimodal perception tasks. Extensive experiments on the nuInsSeg, as well as the large-scale Waymo, show that LWSIS can substantially improve existing weakly supervised segmentation models by only involving 3D data during training. Additionally, LWSIS can also be incorporated into 3D object detectors like PointPainting to boost the 3D detection performance for free. The code and dataset are available at https://github.com/Serenos/LWSIS.

Dongming Wu, Wencheng Han, Yingfei Liu et al.

A new trend in the computer vision community is to capture objects of interest following flexible human command represented by a natural language prompt. However, the progress of using language prompts in driving scenarios is stuck in a bottleneck due to the scarcity of paired prompt-instance data. To address this challenge, we propose the first object-centric language prompt set for driving scenes within 3D, multi-view, and multi-frame space, named NuPrompt. It expands nuScenes dataset by constructing a total of 40,147 language descriptions, each referring to an average of 7.4 object tracklets. Based on the object-text pairs from the new benchmark, we formulate a novel prompt-based driving task, \ie, employing a language prompt to predict the described object trajectory across views and frames. Furthermore, we provide a simple end-to-end baseline model based on Transformer, named PromptTrack. Experiments show that our PromptTrack achieves impressive performance on NuPrompt. We hope this work can provide some new insights for the self-driving community. The data and code have been released at https://github.com/wudongming97/Prompt4Driving.

Dongming Wu, Jiahao Chang, Fan Jia et al.

Topology reasoning aims to comprehensively understand road scenes and present drivable routes in autonomous driving. It requires detecting road centerlines (lane) and traffic elements, further reasoning their topology relationship, i.e., lane-lane topology, and lane-traffic topology. In this work, we first present that the topology score relies heavily on detection performance on lane and traffic elements. Therefore, we introduce a powerful 3D lane detector and an improved 2D traffic element detector to extend the upper limit of topology performance. Further, we propose TopoMLP, a simple yet high-performance pipeline for driving topology reasoning. Based on the impressive detection performance, we develop two simple MLP-based heads for topology generation. TopoMLP achieves state-of-the-art performance on OpenLane-V2 benchmark, i.e., 41.2% OLS with ResNet-50 backbone. It is also the 1st solution for 1st OpenLane Topology in Autonomous Driving Challenge. We hope such simple and strong pipeline can provide some new insights to the community. Code is at https://github.com/wudongming97/TopoMLP.

Junbo Yin, Dingfu Zhou, Liangjun Zhang et al.

Existing approaches for unsupervised point cloud pre-training are constrained to either scene-level or point/voxel-level instance discrimination. Scene-level methods tend to lose local details that are crucial for recognizing the road objects, while point/voxel-level methods inherently suffer from limited receptive field that is incapable of perceiving large objects or context environments. Considering region-level representations are more suitable for 3D object detection, we devise a new unsupervised point cloud pre-training framework, called ProposalContrast, that learns robust 3D representations by contrasting region proposals. Specifically, with an exhaustive set of region proposals sampled from each point cloud, geometric point relations within each proposal are modeled for creating expressive proposal representations. To better accommodate 3D detection properties, ProposalContrast optimizes with both inter-cluster and inter-proposal separation, i.e., sharpening the discriminativeness of proposal representations across semantic classes and object instances. The generalizability and transferability of ProposalContrast are verified on various 3D detectors (i.e., PV-RCNN, CenterPoint, PointPillars and PointRCNN) and datasets (i.e., KITTI, Waymo and ONCE).

Junbo Yin, Jin Fang, Dingfu Zhou et al.

Dominated point cloud-based 3D object detectors in autonomous driving scenarios rely heavily on the huge amount of accurately labeled samples, however, 3D annotation in the point cloud is extremely tedious, expensive and time-consuming. To reduce the dependence on large supervision, semi-supervised learning (SSL) based approaches have been proposed. The Pseudo-Labeling methodology is commonly used for SSL frameworks, however, the low-quality predictions from the teacher model have seriously limited its performance. In this work, we propose a new Pseudo-Labeling framework for semi-supervised 3D object detection, by enhancing the teacher model to a proficient one with several necessary designs. First, to improve the recall of pseudo labels, a Spatialtemporal Ensemble (STE) module is proposed to generate sufficient seed boxes. Second, to improve the precision of recalled boxes, a Clusteringbased Box Voting (CBV) module is designed to get aggregated votes from the clustered seed boxes. This also eliminates the necessity of sophisticated thresholds to select pseudo labels. Furthermore, to reduce the negative influence of wrongly pseudo-labeled samples during the training, a soft supervision signal is proposed by considering Box-wise Contrastive Learning (BCL). The effectiveness of our model is verified on both ONCE and Waymo datasets. For example, on ONCE, our approach significantly improves the baseline by 9.51 mAP. Moreover, with half annotations, our model outperforms the oracle model with full annotations on Waymo.

Junbo Yin, Jianbing Shen, Xin Gao et al.

Previous works for LiDAR-based 3D object detection mainly focus on the single-frame paradigm. In this paper, we propose to detect 3D objects by exploiting temporal information in multiple frames, i.e., the point cloud videos. We empirically categorize the temporal information into short-term and long-term patterns. To encode the short-term data, we present a Grid Message Passing Network (GMPNet), which considers each grid (i.e., the grouped points) as a node and constructs a k-NN graph with the neighbor grids. To update features for a grid, GMPNet iteratively collects information from its neighbors, thus mining the motion cues in grids from nearby frames. To further aggregate the long-term frames, we propose an Attentive Spatiotemporal Transformer GRU (AST-GRU), which contains a Spatial Transformer Attention (STA) module and a Temporal Transformer Attention (TTA) module. STA and TTA enhance the vanilla GRU to focus on small objects and better align the moving objects. Our overall framework supports both online and offline video object detection in point clouds. We implement our algorithm based on prevalent anchor-based and anchor-free detectors. The evaluation results on the challenging nuScenes benchmark show the superior performance of our method, achieving the 1st on the leaderboard without any bells and whistles, by the time the paper is submitted.

Dongming Wu, Tiancai Wang, Yuang Zhang et al.

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

Xingping Dong, Jianbing Shen, Fatih Porikli et al.

In this paper, we provide an intuitive viewing to simplify the Siamese-based trackers by converting the tracking task to a classification. Under this viewing, we perform an in-depth analysis for them through visual simulations and real tracking examples, and find that the failure cases in some challenging situations can be regarded as the issue of missing decisive samples in offline training. Since the samples in the initial (first) frame contain rich sequence-specific information, we can regard them as the decisive samples to represent the whole sequence. To quickly adapt the base model to new scenes, a compact latent network is presented via fully using these decisive samples. Specifically, we present a statistics-based compact latent feature for fast adjustment by efficiently extracting the sequence-specific information. Furthermore, a new diverse sample mining strategy is designed for training to further improve the discrimination ability of the proposed compact latent network. Finally, a conditional updating strategy is proposed to efficiently update the basic models to handle scene variation during the tracking phase. To evaluate the generalization ability and effectiveness and of our method, we apply it to adjust three classical Siamese-based trackers, namely SiamRPN++, SiamFC, and SiamBAN. Extensive experimental results on six recent datasets demonstrate that all three adjusted trackers obtain the superior performance in terms of the accuracy, while having high running speed.

Xingping Dong, Jianbing Shen, Ling Shao

The pioneering method for unsupervised meta-learning, CACTUs, is a clustering-based approach with pseudo-labeling. This approach is model-agnostic and can be combined with supervised algorithms to learn from unlabeled data. However, it often suffers from label inconsistency or limited diversity, which leads to poor performance. In this work, we prove that the core reason for this is lack of a clustering-friendly property in the embedding space. We address this by minimizing the inter- to intra-class similarity ratio to provide clustering-friendly embedding features, and validate our approach through comprehensive experiments. Note that, despite only utilizing a simple clustering algorithm (k-means) in our embedding space to obtain the pseudo-labels, we achieve significant improvement. Moreover, we adopt a progressive evaluation mechanism to obtain more diverse samples in order to further alleviate the limited diversity problem. Finally, our approach is also model-agnostic and can easily be integrated into existing supervised methods. To demonstrate its generalization ability, we integrate it into two representative algorithms: MAML and EP. The results on three main few-shot benchmarks clearly show that the proposed method achieves significant improvement compared to state-of-the-art models. Notably, our approach also outperforms the corresponding supervised method in two tasks.

Chunliang Li, Wencheng Han, Junbo Yin et al.

Concurrent processing of multiple autonomous driving 3D perception tasks within the same spatiotemporal scene poses a significant challenge, in particular due to the computational inefficiencies and feature competition between tasks when using traditional multi-task learning approaches. This paper addresses these issues by proposing a novel unified representation, RepVF, which harmonizes the representation of various perception tasks such as 3D object detection and 3D lane detection within a single framework. RepVF characterizes the structure of different targets in the scene through a vector field, enabling a single-head, multi-task learning model that significantly reduces computational redundancy and feature competition. Building upon RepVF, we introduce RFTR, a network designed to exploit the inherent connections between different tasks by utilizing a hierarchical structure of queries that implicitly model the relationships both between and within tasks. This approach eliminates the need for task-specific heads and parameters, fundamentally reducing the conflicts inherent in traditional multi-task learning paradigms. We validate our approach by combining labels from the OpenLane dataset with the Waymo Open dataset. Our work presents a significant advancement in the efficiency and effectiveness of multi-task perception in autonomous driving, offering a new perspective on handling multiple 3D perception tasks synchronously and in parallel. The code will be available at: https://github.com/jbji/RepVF

Jiawei Liu, Xingping Dong, Sanyuan Zhao et al.

Recent years have witnessed huge successes in 3D object detection to recognize common objects for autonomous driving (e.g., vehicles and pedestrians). However, most methods rely heavily on a large amount of well-labeled training data. This limits their capability of detecting rare fine-grained objects (e.g., police cars and ambulances), which is important for special cases, such as emergency rescue, and so on. To achieve simultaneous detection for both common and rare objects, we propose a novel task, called generalized few-shot 3D object detection, where we have a large amount of training data for common (base) objects, but only a few data for rare (novel) classes. Specifically, we analyze in-depth differences between images and point clouds, and then present a practical principle for the few-shot setting in the 3D LiDAR dataset. To solve this task, we propose a simple and effective detection framework, including (1) an incremental fine-tuning method to extend existing 3D detection models to recognize both common and rare objects, and (2) a sample adaptive balance loss to alleviate the issue of long-tailed data distribution in autonomous driving scenarios. On the nuScenes dataset, we conduct sufficient experiments to demonstrate that our approach can successfully detect the rare (novel) classes that contain only a few training data, while also maintaining the detection accuracy of common objects.

Yukun Zhai, Xiaoqiang Zhang, Xiameng Qin et al.

End-to-end text spotting is a vital computer vision task that aims to integrate scene text detection and recognition into a unified framework. Typical methods heavily rely on Region-of-Interest (RoI) operations to extract local features and complex post-processing steps to produce final predictions. To address these limitations, we propose TextFormer, a query-based end-to-end text spotter with Transformer architecture. Specifically, using query embedding per text instance, TextFormer builds upon an image encoder and a text decoder to learn a joint semantic understanding for multi-task modeling. It allows for mutual training and optimization of classification, segmentation, and recognition branches, resulting in deeper feature sharing without sacrificing flexibility or simplicity. Additionally, we design an Adaptive Global aGgregation (AGG) module to transfer global features into sequential features for reading arbitrarily-shaped texts, which overcomes the sub-optimization problem of RoI operations. Furthermore, potential corpus information is utilized from weak annotations to full labels through mixed supervision, further improving text detection and end-to-end text spotting results. Extensive experiments on various bilingual (i.e., English and Chinese) benchmarks demonstrate the superiority of our method. Especially on TDA-ReCTS dataset, TextFormer surpasses the state-of-the-art method in terms of 1-NED by 13.2%.

Yucheng Zhou, Xiubo Geng, Tao Shen et al.

Large Language Models (LLMs) have ushered in a transformative era in the field of natural language processing, excelling in tasks related to text comprehension and generation. Nevertheless, they encounter difficulties when confronted with chaotic contexts (e.g., distractors rather than long irrelevant context), leading to the inadvertent omission of certain details within the chaotic context. In response to these challenges, we introduce the "Thread of Thought" (ThoT) strategy, which draws inspiration from human cognitive processes. ThoT systematically segments and analyzes extended contexts while adeptly selecting pertinent information. This strategy serves as a versatile "plug-and-play" module, seamlessly integrating with various LLMs and prompting techniques. In the experiments, we utilize the PopQA and EntityQ datasets, as well as a Multi-Turn Conversation Response dataset (MTCR) we collected, to illustrate that ThoT significantly improves reasoning performance compared to other prompting techniques.

Hongji Yang, Songlian Li, Yucheng Zhou et al.

Recent diffusion models achieve strong photorealism and fluency in video generation, yet remain fragile under abstract, sparse or complex conditions, leading to poor performance in professional production workflows such as storyboard sketches and clay render conditions. Existing video generation models, either inject conditions through adapters or couple a generic vision-language model (VLM) within a diffusion backbone, leaving a capability gap and failing to produce the videos that align with the user's creative intent. We present CogOmniControl, a reasoning-driven framework that factorizes controllable video generation into creative intent cognition and generation. Specifically, we train a specialized CogVLM using authentic anime production data. Compared to generic VLMs, it generates more professional and clear outputs, accurately cognizing user creative intent from sparse and abstract conditions and tuning these cues into dense reasoning output. Besides, CogOmniDiT unifies the controls from various conditions through in-context generation and is aligned to the CogVLM reasoning outputs via reinforcement learning. Furthermore, leveraging CogVLM's robust capability in guiding video generation, we release its potential in planning specific evaluators and enable a Best-of-N selection for the generated videos. This integration transforms the entire framework into a closed-loop "harness-like" architecture. We further introduce CogReasonBench and CogControlBench, built from professional workflows data that carry genuine creative intent rather than simulated ones. Experiments on two benchmarks show that CogOmniControl surpassed the existing open-source models. The project website: https://um-lab.github.io/CogOmniControl/

Wencheng Han, Junbo Yin, Jianbing Shen

Monocular depth estimation is known as an ill-posed task in which objects in a 2D image usually do not contain sufficient information to predict their depth. Thus, it acts differently from other tasks (e.g., classification and segmentation) in many ways. In this paper, we find that self-supervised monocular depth estimation shows a direction sensitivity and environmental dependency in the feature representation. But the current backbones borrowed from other tasks pay less attention to handling different types of environmental information, limiting the overall depth accuracy. To bridge this gap, we propose a new Direction-aware Cumulative Convolution Network (DaCCN), which improves the depth feature representation in two aspects. First, we propose a direction-aware module, which can learn to adjust the feature extraction in each direction, facilitating the encoding of different types of information. Secondly, we design a new cumulative convolution to improve the efficiency for aggregating important environmental information. Experiments show that our method achieves significant improvements on three widely used benchmarks, KITTI, Cityscapes, and Make3D, setting a new state-of-the-art performance on the popular benchmarks with all three types of self-supervision.

Zhuding Liang, Tianyi Yan, Dubing Chen et al.

Generative world models increasingly rely on 4D occupancy for realistic autonomous driving simulation. However, existing generation frameworks depend on rigid geometric conditions (e.g., explicit trajectories) or simplistic attribute-level text, failing to orchestrate complex, sequential multi-agent interactions. To address this semantic-spatiotemporal gap, we propose OccDirector, a pioneering framework that generates 4D occupancy dynamics conditioned solely on natural language. Operating as a ``scenario director'', OccDirector maps natural language scripts into physically plausible voxel dynamics without requiring geometric priors. Technically, it employs a VLM-driven Spatio-Temporal MMDiT equipped with a history-prefix anchoring strategy to ensure long-horizon interaction consistency. Furthermore, we introduce OccInteract-85k, a novel dataset uniquely annotated with multi-level language instructions: ranging from static layouts to intricate multi-agent behaviors, alongside a novel VLM-based evaluation benchmark. Extensive experiments demonstrate that OccDirector achieves state-of-the-art generation quality and unprecedented instruction-following capabilities, successfully shifting the paradigm from appearance synthesis to language-driven behavior orchestration.

Junbo Yin, Jianbing Shen, Runnan Chen et al.

Bird's eye view (BEV) representation has emerged as a dominant solution for describing 3D space in autonomous driving scenarios. However, objects in the BEV representation typically exhibit small sizes, and the associated point cloud context is inherently sparse, which leads to great challenges for reliable 3D perception. In this paper, we propose IS-Fusion, an innovative multimodal fusion framework that jointly captures the Instance- and Scene-level contextual information. IS-Fusion essentially differs from existing approaches that only focus on the BEV scene-level fusion by explicitly incorporating instance-level multimodal information, thus facilitating the instance-centric tasks like 3D object detection. It comprises a Hierarchical Scene Fusion (HSF) module and an Instance-Guided Fusion (IGF) module. HSF applies Point-to-Grid and Grid-to-Region transformers to capture the multimodal scene context at different granularities. IGF mines instance candidates, explores their relationships, and aggregates the local multimodal context for each instance. These instances then serve as guidance to enhance the scene feature and yield an instance-aware BEV representation. On the challenging nuScenes benchmark, IS-Fusion outperforms all the published multimodal works to date. Code is available at: https://github.com/yinjunbo/IS-Fusion.

Wencheng Han, Jianbing Shen

The curve-based lane representation is a popular approach in many lane detection methods, as it allows for the representation of lanes as a whole object and maximizes the use of holistic information about the lanes. However, the curves produced by these methods may not fit well with irregular lines, which can lead to gaps in performance compared to indirect representations such as segmentation-based or point-based methods. We have observed that these lanes are not intended to be irregular, but they appear zigzagged in the perspective view due to being drawn on uneven pavement. In this paper, we propose a new approach to the lane detection task by decomposing it into two parts: curve modeling and ground height regression. Specifically, we use a parameterized curve to represent lanes in the BEV space to reflect the original distribution of lanes. For the second part, since ground heights are determined by natural factors such as road conditions and are less holistic, we regress the ground heights of key points separately from the curve modeling. Additionally, we have unified the 2D and 3D lane detection tasks by designing a new framework and a series of losses to guide the optimization of models with or without 3D lane labels. Our experiments on 2D lane detection benchmarks (TuSimple and CULane), as well as the recently proposed 3D lane detection datasets (ONCE-3Dlane and OpenLane), have shown significant improvements. We will make our well-documented source code publicly available.

Li Xiang, Junbo Yin, Wei Li et al.

Vehicle-to-Everything (V2X) collaborative perception has recently gained significant attention due to its capability to enhance scene understanding by integrating information from various agents, e.g., vehicles, and infrastructure. However, current works often treat the information from each agent equally, ignoring the inherent domain gap caused by the utilization of different LiDAR sensors of each agent, thus leading to suboptimal performance. In this paper, we propose DI-V2X, that aims to learn Domain-Invariant representations through a new distillation framework to mitigate the domain discrepancy in the context of V2X 3D object detection. DI-V2X comprises three essential components: a domain-mixing instance augmentation (DMA) module, a progressive domain-invariant distillation (PDD) module, and a domain-adaptive fusion (DAF) module. Specifically, DMA builds a domain-mixing 3D instance bank for the teacher and student models during training, resulting in aligned data representation. Next, PDD encourages the student models from different domains to gradually learn a domain-invariant feature representation towards the teacher, where the overlapping regions between agents are employed as guidance to facilitate the distillation process. Furthermore, DAF closes the domain gap between the students by incorporating calibration-aware domain-adaptive attention. Extensive experiments on the challenging DAIR-V2X and V2XSet benchmark datasets demonstrate DI-V2X achieves remarkable performance, outperforming all the previous V2X models. Code is available at https://github.com/Serenos/DI-V2X

Wencheng Han, Jianbing Shen

In the area of self-supervised monocular depth estimation, models that utilize rich-resource inputs, such as high-resolution and multi-frame inputs, typically achieve better performance than models that use ordinary single image input. However, these rich-resource inputs may not always be available, limiting the applicability of these methods in general scenarios. In this paper, we propose Rich-resource Prior Depth estimator (RPrDepth), which only requires single input image during the inference phase but can still produce highly accurate depth estimations comparable to rich resource based methods. Specifically, we treat rich-resource data as prior information and extract features from it as reference features in an offline manner. When estimating the depth for a single-image image, we search for similar pixels from the rich-resource features and use them as prior information to estimate the depth. Experimental results demonstrate that our model outperform other single-image model and can achieve comparable or even better performance than models with rich-resource inputs, only using low-resolution single-image input.

Lingran Song, Yucheng Zhou, Jianbing Shen

Despite significant progress in pixel-level medical image analysis, existing medical image segmentation models rarely explore medical segmentation and diagnosis tasks jointly. However, it is crucial for patients that models can provide explainable diagnoses along with medical segmentation results. In this paper, we introduce a medical vision-language task named Medical Diagnosis Segmentation (MDS), which aims to understand clinical queries for medical images and generate the corresponding segmentation masks as well as diagnostic results. To facilitate this task, we first present the Multimodal Multi-disease Medical Diagnosis Segmentation (M3DS) dataset, containing diverse multimodal multi-disease medical images paired with their corresponding segmentation masks and diagnosis chain-of-thought, created via an automated diagnosis chain-of-thought generation pipeline. Moreover, we propose Sim4Seg, a novel framework that improves the performance of diagnosis segmentation by taking advantage of the Region-Aware Vision-Language Similarity to Mask (RVLS2M) module. To improve overall performance, we investigate a test-time scaling strategy for MDS tasks. Experimental results demonstrate that our method outperforms the baselines in both segmentation and diagnosis.

Yucheng Zhou, Lingran Song, Jianbing Shen

Recent advancements in medical Large Language Models (LLMs) have showcased their powerful reasoning and diagnostic capabilities. Despite their success, current unified multimodal medical LLMs face limitations in knowledge update costs, comprehensiveness, and flexibility. To address these challenges, we introduce the Modular Multi-Agent Framework for Multi-Modal Medical Diagnosis (MAM). Inspired by our empirical findings highlighting the benefits of role assignment and diagnostic discernment in LLMs, MAM decomposes the medical diagnostic process into specialized roles: a General Practitioner, Specialist Team, Radiologist, Medical Assistant, and Director, each embodied by an LLM-based agent. This modular and collaborative framework enables efficient knowledge updates and leverages existing medical LLMs and knowledge bases. Extensive experimental evaluations conducted on a wide range of publicly accessible multimodal medical datasets, incorporating text, image, audio, and video modalities, demonstrate that MAM consistently surpasses the performance of modality-specific LLMs. Notably, MAM achieves significant performance improvements ranging from 18% to 365% compared to baseline models. Our code is released at https://github.com/yczhou001/MAM.

Dubing Chen, Wencheng Han, Jin Fang et al.

In this technical report, we present our solution for the Vision-Centric 3D Occupancy and Flow Prediction track in the nuScenes Open-Occ Dataset Challenge at CVPR 2024. Our innovative approach involves a dual-stage framework that enhances 3D occupancy and flow predictions by incorporating adaptive forward view transformation and flow modeling. Initially, we independently train the occupancy model, followed by flow prediction using sequential frame integration. Our method combines regression with classification to address scale variations in different scenes, and leverages predicted flow to warp current voxel features to future frames, guided by future frame ground truth. Experimental results on the nuScenes dataset demonstrate significant improvements in accuracy and robustness, showcasing the effectiveness of our approach in real-world scenarios. Our single model based on Swin-Base ranks second on the public leaderboard, validating the potential of our method in advancing autonomous car perception systems.

Tianyi Yan, Junbo Yin, Xianpeng Lang et al.

To enhance autonomous driving safety in complex scenarios, various methods have been proposed to simulate LiDAR point cloud data. Nevertheless, these methods often face challenges in producing high-quality, diverse, and controllable foreground objects. To address the needs of object-aware tasks in 3D perception, we introduce OLiDM, a novel framework capable of generating high-fidelity LiDAR data at both the object and the scene levels. OLiDM consists of two pivotal components: the Object-Scene Progressive Generation (OPG) module and the Object Semantic Alignment (OSA) module. OPG adapts to user-specific prompts to generate desired foreground objects, which are subsequently employed as conditions in scene generation, ensuring controllable outputs at both the object and scene levels. This also facilitates the association of user-defined object-level annotations with the generated LiDAR scenes. Moreover, OSA aims to rectify the misalignment between foreground objects and background scenes, enhancing the overall quality of the generated objects. The broad effectiveness of OLiDM is demonstrated across various LiDAR generation tasks, as well as in 3D perception tasks. Specifically, on the KITTI-360 dataset, OLiDM surpasses prior state-of-the-art methods such as UltraLiDAR by 17.5 in FPD. Additionally, in sparse-to-dense LiDAR completion, OLiDM achieves a significant improvement over LiDARGen, with a 57.47\% increase in semantic IoU. Moreover, OLiDM enhances the performance of mainstream 3D detectors by 2.4\% in mAP and 1.9\% in NDS, underscoring its potential in advancing object-aware 3D tasks. Code is available at: https://yanty123.github.io/OLiDM.

Xingtai Gui, Meijie Zhang, Tianyi Yan et al.

End-to-end autonomous driving aims to generate safe and plausible planning policies from raw sensor input. Driving world models have shown great potential in learning rich representations by predicting the future evolution of a driving scene. However, existing driving world models primarily focus on visual scene representation, and motion representation is not explicitly designed to be planner-shared and inheritable, leaving a schism between the optimization of visual scene generation and the requirements of precise motion planning. We present WorldDrive, a holistic framework that couples scene generation and real-time planning via unifying vision and motion representation. We first introduce a Trajectory-aware Driving World Model, which conditions on a trajectory vocabulary to enforce consistency between visual dynamics and motion intentions, enabling the generation of diverse and plausible future scenes conditioned on a specific trajectory. We transfer the vision and motion encoders to a downstream Multi-modal Planner, ensuring the driving policy operates on mature representations pre-optimized by scene generation. A simple interaction between motion representation, visual representation, and ego status can generate high-quality, multi-modal trajectories. Furthermore, to exploit the world model's foresight, we propose a Future-aware Rewarder, which distills future latent representation from the frozen world model to evaluate and select optimal trajectories in real-time. Extensive experiments on the NAVSIM, NAVSIM-v2, and nuScenes benchmarks demonstrate that WorldDrive achieves leading planning performance among vision-only methods while maintaining high-fidelity action-controlled video generation capabilities, providing strong evidence for the effectiveness of unifying vision and motion representation for robust autonomous driving.

Huan Zheng, Wencheng Han, Jianbing Shen

Recovering high-quality depth maps from compressed sources has gained significant attention due to the limitations of consumer-grade depth cameras and the bandwidth restrictions during data transmission. However, current methods still suffer from two challenges. First, bit-depth compression produces a uniform depth representation in regions with subtle variations, hindering the recovery of detailed information. Second, densely distributed random noise reduces the accuracy of estimating the global geometric structure of the scene. To address these challenges, we propose a novel framework, termed geometry-decoupled network (GDNet), for compressed depth map super-resolution that decouples the high-quality depth map reconstruction process by handling global and detailed geometric features separately. To be specific, we propose the fine geometry detail encoder (FGDE), which is designed to aggregate fine geometry details in high-resolution low-level image features while simultaneously enriching them with complementary information from low-resolution context-level image features. In addition, we develop the global geometry encoder (GGE) that aims at suppressing noise and extracting global geometric information effectively via constructing compact feature representation in a low-rank space. We conduct experiments on multiple benchmark datasets, demonstrating that our GDNet significantly outperforms current methods in terms of geometric consistency and detail recovery. In the ECCV 2024 AIM Compressed Depth Upsampling Challenge, our solution won the 1st place award. Our codes are available at: https://github.com/Ian0926/GDNet.

Hongji Yang, Yucheng Zhou, Wencheng Han et al.

While Large Vision-Language Models (VLMs) demonstrate impressive general visual capabilities, they remain artistically blind and unable to offer professional evaluation of artworks within specific artistic domains like human experts. To bridge this gap, we transform VLMs into experts capable of professional-grade painting evaluation in the Chinese Artistic Domain, which is more abstract and demands extensive artistic training for evaluation. We introduce HanMo-Bench, a new dataset that features authentic auction-grade masterpieces and AI-generated works, grounded in real-world market valuations. To realize the rigorous judgment, we propose the HanMoVLM and construct a Chain-of-Thought (CoT) validated by experts. This CoT guides the model to perform expert-level reasoning: from content identification and Region of Interest (RoI) localization to professional evaluation, guided by both theme-specific evaluation and typical three-tier evaluation in Chinese paintings. Furthermore, we design a reward function to refine the reasoning process of the HanMoVLM to improve the accuracy. We demonstrate that HanMoVLM can serve as a critical backbone for Test-time Scaling in image generation. By acting as a high-quality verifier, HanMoVLM enables generative models to select the most artistically superior outputs from multiple candidates. Experimental results and human studies confirm that the proposed HanMoVLM effectively bridges the gap, achieving a high consistency with professional experts and significantly improving the quality of Chinese Painting generation.

Xingtai Gui, Jianbo Zhao, Wencheng Han et al.

End-to-end autonomous driving systems directly generate driving policies from raw sensor inputs. While these systems can extract effective environmental features for planning, relying on auxiliary perception tasks, developing perception annotation-free planning paradigms has become increasingly critical due to the high cost of manual perception annotation. In this work, we propose TrajDiff, a Trajectory-oriented BEV Conditioned Diffusion framework that establishes a fully perception annotation-free generative method for end-to-end autonomous driving. TrajDiff requires only raw sensor inputs and future trajectory, constructing Gaussian BEV heatmap targets that inherently capture driving modalities. We design a simple yet effective trajectory-oriented BEV encoder to extract the TrajBEV feature without perceptual supervision. Furthermore, we introduce Trajectory-oriented BEV Diffusion Transformer (TB-DiT), which leverages ego-state information and the predicted TrajBEV features to directly generate diverse yet plausible trajectories, eliminating the need for handcrafted motion priors. Beyond architectural innovations, TrajDiff enables exploration of data scaling benefits in the annotation-free setting. Evaluated on the NAVSIM benchmark, TrajDiff achieves 87.5 PDMS, establishing state-of-the-art performance among all annotation-free methods. With data scaling, it further improves to 88.5 PDMS, which is comparable to advanced perception-based approaches. Our code and model will be made publicly available.

Huan Zheng, Yucheng Zhou, Tianyi Yan et al.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable potential in medical image analysis. However, their application in gastrointestinal endoscopy is currently hindered by two critical limitations: the misalignment between general model reasoning and standardized clinical cognitive pathways, and the lack of causal association between visual features and diagnostic outcomes. In this paper, we propose a novel Clinical-Cognitive-Aligned (CogAlign) framework to address these challenges. First, we endow the model with rigorous clinical analytical capabilities by constructing the hierarchical clinical cognition dataset and employing Supervised Fine-Tuning (SFT). Unlike conventional approaches, this strategy internalizes the hierarchical diagnostic logic of experts, ranging from anatomical localization and morphological evaluation to microvascular analysis, directly into the model. Second, to eliminate visual bias, we provide a theoretical analysis demonstrating that standard supervised tuning inevitably converges to spurious background correlations. Guided by this insight, we propose a counterfactual-driven reinforcement learning strategy to enforce causal rectification. By generating counterfactual normal samples via lesion masking and optimizing through clinical-cognition-centric rewards, we constrain the model to strictly ground its diagnosis in causal lesion features. Extensive experiments demonstrate that our approach achieves State-of-the-Art (SoTA) performance across multiple benchmarks, significantly enhancing diagnostic accuracy in complex clinical scenarios. All source code and datasets will be made publicly available.

Wencheng Han, Dongqian Guo, Xiao Chen et al.

Metal artifacts in CT slices have long posed challenges in medical diagnostics. These artifacts degrade image quality, resulting in suboptimal visualization and complicating the accurate interpretation of tissues adjacent to metal implants. To address these issues, we introduce the Latent Gemstone Spectral Imaging (GSI) Alignment Framework, which effectively reduces metal artifacts while avoiding the introduction of noise information. Our work is based on a key finding that even artifact-affected ordinary CT sequences contain sufficient information to discern detailed structures. The challenge lies in the inability to clearly represent this information. To address this issue, we developed an Alignment Framework that adjusts the representation of ordinary CT images to match GSI CT sequences. GSI is an advanced imaging technique using multiple energy levels to mitigate artifacts caused by metal implants. By aligning the representation to GSI data, we can effectively suppress metal artifacts while clearly revealing detailed structure, without introducing extraneous information into CT sequences. To facilitate the application, we propose a new dataset, Artifacts-GSI, captured from real patients with metal implants, and establish a new benchmark based on this dataset. Experimental results show that our method significantly reduces metal artifacts and greatly enhances the readability of CT slices. All our code and data are available at: https://um-lab.github.io/GSI-MAR/

Qinghao Meng, Chenming Wu, Liangjun Zhang et al.

3D object detection is essential in autonomous driving, providing vital information about moving objects and obstacles. Detecting objects in distant regions with only a few LiDAR points is still a challenge, and numerous strategies have been developed to address point cloud sparsity through densification.This paper presents a joint completion and detection framework that improves the detection feature in sparse areas while maintaining costs unchanged. Specifically, we propose TransBridge, a novel transformer-based up-sampling block that fuses the features from the detection and completion networks.The detection network can benefit from acquiring implicit completion features derived from the completion network. Additionally, we design the Dynamic-Static Reconstruction (DSRecon) module to produce dense LiDAR data for the completion network, meeting the requirement for dense point cloud ground truth.Furthermore, we employ the transformer mechanism to establish connections between channels and spatial relations, resulting in a high-resolution feature map used for completion purposes.Extensive experiments on the nuScenes and Waymo datasets demonstrate the effectiveness of the proposed framework.The results show that our framework consistently improves end-to-end 3D object detection, with the mean average precision (mAP) ranging from 0.7 to 1.5 across multiple methods, indicating its generalization ability. For the two-stage detection framework, it also boosts the mAP up to 5.78 points.

Hongjun Chen, Huan Zheng, Wencheng Han et al.

Existing video-based 3D Human Mesh Recovery (HMR) methods often produce physically implausible results, stemming from their reliance on flawed intermediate 3D pose anchors and their inability to effectively model complex spatiotemporal dynamics. To overcome these deep-rooted architectural problems, we introduce HMRMamba, a new paradigm for HMR that pioneers the use of Structured State Space Models (SSMs) for their efficiency and long-range modeling prowess. Our framework is distinguished by two core contributions. First, the Geometry-Aware Lifting Module, featuring a novel dual-scan Mamba architecture, creates a robust foundation for reconstruction. It directly grounds the 2D-to-3D pose lifting process with geometric cues from image features, producing a highly reliable 3D pose sequence that serves as a stable anchor. Second, the Motion-guided Reconstruction Network leverages this anchor to explicitly process kinematic patterns over time. By injecting this crucial temporal awareness, it significantly enhances the final mesh's coherence and robustness, particularly under occlusion and motion blur. Comprehensive evaluations on 3DPW, MPI-INF-3DHP, and Human3.6M benchmarks confirm that HMRMamba sets a new state-of-the-art, outperforming existing methods in both reconstruction accuracy and temporal consistency while offering superior computational efficiency.

Yucheng Zhou, Xiang Li, Qianning Wang et al.

In Large Visual Language Models (LVLMs), the efficacy of In-Context Learning (ICL) remains limited by challenges in cross-modal interactions and representation disparities. To overcome these challenges, we introduce a novel Visual In-Context Learning (VICL) method comprising Visual Demonstration Retrieval, Intent-Oriented Image Summarization, and Intent-Oriented Demonstration Composition. Our approach retrieves images via ''Retrieval & Rerank'' paradigm, summarises images with task intent and task-specific visual parsing, and composes language-based demonstrations that reduce token count and alleviate cross-modal interaction problem. Experimental evaluations on five visual reasoning datasets demonstrate the effectiveness of our method. Moreover, our extensive experiments leverage information flow analysis to elucidate the effectiveness of our method, and investigate the impact of length and position of demonstrations for LVLM. The use of in-context unlearning further shows promise in resetting specific model knowledge without retraining.

Dongming Wu, Yanping Fu, Saike Huang et al.

General robotic grasping systems require accurate object affordance perception in diverse open-world scenarios following human instructions. However, current studies suffer from the problem of lacking reasoning-based large-scale affordance prediction data, leading to considerable concern about open-world effectiveness. To address this limitation, we build a large-scale grasping-oriented affordance segmentation benchmark with human-like instructions, named RAGNet. It contains 273k images, 180 categories, and 26k reasoning instructions. The images cover diverse embodied data domains, such as wild, robot, ego-centric, and even simulation data. They are carefully annotated with an affordance map, while the difficulty of language instructions is largely increased by removing their category name and only providing functional descriptions. Furthermore, we propose a comprehensive affordance-based grasping framework, named AffordanceNet, which consists of a VLM pre-trained on our massive affordance data and a grasping network that conditions an affordance map to grasp the target. Extensive experiments on affordance segmentation benchmarks and real-robot manipulation tasks show that our model has a powerful open-world generalization ability. Our data and code is available at https://github.com/wudongming97/AffordanceNet.

Tao Zhou, Deng-Ping Fan, Ming-Ming Cheng et al.

Salient object detection (SOD), which simulates the human visual perception system to locate the most attractive object(s) in a scene, has been widely applied to various computer vision tasks. Now, with the advent of depth sensors, depth maps with affluent spatial information that can be beneficial in boosting the performance of SOD, can easily be captured. Although various RGB-D based SOD models with promising performance have been proposed over the past several years, an in-depth understanding of these models and challenges in this topic remains lacking. In this paper, we provide a comprehensive survey of RGB-D based SOD models from various perspectives, and review related benchmark datasets in detail. Further, considering that the light field can also provide depth maps, we review SOD models and popular benchmark datasets from this domain as well. Moreover, to investigate the SOD ability of existing models, we carry out a comprehensive evaluation, as well as attribute-based evaluation of several representative RGB-D based SOD models. Finally, we discuss several challenges and open directions of RGB-D based SOD for future research. All collected models, benchmark datasets, source code links, datasets constructed for attribute-based evaluation, and codes for evaluation will be made publicly available at https://github.com/taozh2017/RGBDSODsurvey

Zhijie Zhang, Huazhu Fu, Hang Dai et al.

Segmentation is a fundamental task in medical image analysis. However, most existing methods focus on primary region extraction and ignore edge information, which is useful for obtaining accurate segmentation. In this paper, we propose a generic medical segmentation method, called Edge-aTtention guidance Network (ET-Net), which embeds edge-attention representations to guide the segmentation network. Specifically, an edge guidance module is utilized to learn the edge-attention representations in the early encoding layers, which are then transferred to the multi-scale decoding layers, fused using a weighted aggregation module. The experimental results on four segmentation tasks (i.e., optic disc/cup and vessel segmentation in retinal images, and lung segmentation in chest X-Ray and CT images) demonstrate that preserving edge-attention representations contributes to the final segmentation accuracy, and our proposed method outperforms current state-of-the-art segmentation methods. The source code of our method is available at https://github.com/ZzzJzzZ/ETNet.

Aamir Mustafa, Salman H. Khan, Munawar Hayat et al.

Convolutional Neural Networks have achieved significant success across multiple computer vision tasks. However, they are vulnerable to carefully crafted, human-imperceptible adversarial noise patterns which constrain their deployment in critical security-sensitive systems. This paper proposes a computationally efficient image enhancement approach that provides a strong defense mechanism to effectively mitigate the effect of such adversarial perturbations. We show that deep image restoration networks learn mapping functions that can bring off-the-manifold adversarial samples onto the natural image manifold, thus restoring classification towards correct classes. A distinguishing feature of our approach is that, in addition to providing robustness against attacks, it simultaneously enhances image quality and retains models performance on clean images. Furthermore, the proposed method does not modify the classifier or requires a separate mechanism to detect adversarial images. The effectiveness of the scheme has been demonstrated through extensive experiments, where it has proven a strong defense in gray-box settings. The proposed scheme is simple and has the following advantages: (1) it does not require any model training or parameter optimization, (2) it complements other existing defense mechanisms, (3) it is agnostic to the attacked model and attack type and (4) it provides superior performance across all popular attack algorithms. Our codes are publicly available at https://github.com/aamir-mustafa/super-resolution-adversarial-defense.

Siyuan Qi, Wenguan Wang, Baoxiong Jia et al.

This paper addresses the task of detecting and recognizing human-object interactions (HOI) in images and videos. We introduce the Graph Parsing Neural Network (GPNN), a framework that incorporates structural knowledge while being differentiable end-to-end. For a given scene, GPNN infers a parse graph that includes i) the HOI graph structure represented by an adjacency matrix, and ii) the node labels. Within a message passing inference framework, GPNN iteratively computes the adjacency matrices and node labels. We extensively evaluate our model on three HOI detection benchmarks on images and videos: HICO-DET, V-COCO, and CAD-120 datasets. Our approach significantly outperforms state-of-art methods, verifying that GPNN is scalable to large datasets and applies to spatial-temporal settings. The code is available at https://github.com/SiyuanQi/gpnn.

Wencheng Han, Dongqian Guo, Cheng-Zhong Xu et al.

In the field of autonomous driving, two important features of autonomous driving car systems are the explainability of decision logic and the accuracy of environmental perception. This paper introduces DME-Driver, a new autonomous driving system that enhances the performance and reliability of autonomous driving system. DME-Driver utilizes a powerful vision language model as the decision-maker and a planning-oriented perception model as the control signal generator. To ensure explainable and reliable driving decisions, the logical decision-maker is constructed based on a large vision language model. This model follows the logic employed by experienced human drivers and makes decisions in a similar manner. On the other hand, the generation of accurate control signals relies on precise and detailed environmental perception, which is where 3D scene perception models excel. Therefore, a planning oriented perception model is employed as the signal generator. It translates the logical decisions made by the decision-maker into accurate control signals for the self-driving cars. To effectively train the proposed model, a new dataset for autonomous driving was created. This dataset encompasses a diverse range of human driver behaviors and their underlying motivations. By leveraging this dataset, our model achieves high-precision planning accuracy through a logical thinking process.

Yucheng Zhou, Lingran Song, Jianbing Shen

Existing Medical Large Vision-Language Models (Med-LVLMs), encapsulating extensive medical knowledge, demonstrate excellent capabilities in understanding medical images. However, there remain challenges in visual localization in medical images, which is crucial for abnormality detection and interpretation. To address these issues, we propose a novel UMed-LVLM designed to unveil medical abnormalities. Specifically, we collect a Medical Abnormalities Unveiling (MAU) dataset and propose a two-stage training method for UMed-LVLM training. To collect MAU dataset, we propose a prompt method utilizing the GPT-4V to generate diagnoses based on identified abnormal areas in medical images. Moreover, the two-stage training method includes Abnormal-Aware Instruction Tuning and Abnormal-Aware Rewarding, comprising Relevance Reward, Abnormal Localization Reward and Vision Relevance Reward. Experimental results demonstrate that our UMed-LVLM significantly outperforms existing Med-LVLMs in identifying and understanding medical abnormalities, achieving a 58% improvement over the baseline. In addition, this work shows that enhancing the abnormality detection capabilities of Med-LVLMs significantly improves their understanding of medical images and generalization capability.

Yucheng Zhou, Zhi Rao, Jun Wan et al.

Large Vision-Language Models (LVLMs) excel in cross-model tasks but experience performance declines in long-context reasoning due to overreliance on textual information and reduced visual dependency. In this study, we empirically analyze LVLMs in long-context reasoning, revealing that increased context length leads to a higher dependence on language at the expense of visual dependency. To address this issue, we propose a novel training-free context pruning method that selectively removes less critical textual information. Our approach enhances visual dependency and reduces textual noise, thereby improving LVLM performance in long-context reasoning. We validate our method by constructing a long-context dataset, demonstrating its effectiveness across various LVLMs. Moreover, further analysis confirms the robustness of different token pruning strategies and preliminary explores scaling laws between pruning rates and context length.

Tianyi Yan, Dongming Wu, Wencheng Han et al.

Autonomous driving evaluation requires simulation environments that closely replicate actual road conditions, including real-world sensory data and responsive feedback loops. However, many existing simulations need to predict waypoints along fixed routes on public datasets or synthetic photorealistic data, \ie, open-loop simulation usually lacks the ability to assess dynamic decision-making. While the recent efforts of closed-loop simulation offer feedback-driven environments, they cannot process visual sensor inputs or produce outputs that differ from real-world data. To address these challenges, we propose DrivingSphere, a realistic and closed-loop simulation framework. Its core idea is to build 4D world representation and generate real-life and controllable driving scenarios. In specific, our framework includes a Dynamic Environment Composition module that constructs a detailed 4D driving world with a format of occupancy equipping with static backgrounds and dynamic objects, and a Visual Scene Synthesis module that transforms this data into high-fidelity, multi-view video outputs, ensuring spatial and temporal consistency. By providing a dynamic and realistic simulation environment, DrivingSphere enables comprehensive testing and validation of autonomous driving algorithms, ultimately advancing the development of more reliable autonomous cars. The benchmark will be publicly released.

Chenglin Wang, Yucheng Zhou, Zijie Zhai et al.

Protein inverse folding is a fundamental problem in bioinformatics, aiming to recover the amino acid sequences from a given protein backbone structure. Despite the success of existing methods, they struggle to fully capture the intricate inter-residue relationships critical for accurate sequence prediction. We propose a novel method that leverages diffusion models with representation alignment (DMRA), which enhances diffusion-based inverse folding by (1) proposing a shared center that aggregates contextual information from the entire protein structure and selectively distributes it to each residue; and (2) aligning noisy hidden representations with clean semantic representations during the denoising process. This is achieved by predefined semantic representations for amino acid types and a representation alignment method that utilizes type embeddings as semantic feedback to normalize each residue. In experiments, we conduct extensive evaluations on the CATH4.2 dataset to demonstrate that DMRA outperforms leading methods, achieving state-of-the-art performance and exhibiting strong generalization capabilities on the TS50 and TS500 datasets.

Han Song, Yucheng Zhou, Jianbing Shen et al.

Combining Chain-of-Thought (CoT) with Reinforcement Learning (RL) improves text-to-image (T2I) generation, yet the underlying interaction between CoT's exploration and RL's optimization remains unclear. We present a systematic entropy-based analysis that yields three key insights: (1) CoT expands the generative exploration space, while RL contracts it toward high-reward regions; (2) final reward is strongly negatively correlated with both the mean and variance of image-token entropy, highlighting the need to reduce uncertainty and instability; and (3) the entropy of the textual CoT directly governs downstream image quality, with lower-entropy CoTs leading to better generations. Motivated by these findings, we propose Entropy-Guided Group Relative Policy Optimization (EG-GRPO), a fine-tuning strategy that reallocates optimization budget by uncertainty: low-entropy tokens are excluded from reward-driven updates to preserve stability, while high-entropy tokens receive an entropy bonus that encourages structured exploration without collapse. Experiments on standard T2I benchmarks demonstrate that EG-GRPO achieves state-of-the-art performance.

Yucheng Zhou, Jianbing Shen

Autoregressive models have shown superior performance and efficiency in image generation, but remain constrained by high computational costs and prolonged training times in video generation. In this study, we explore methods to accelerate training for autoregressive video generation models through empirical analyses. Our results reveal that while training on fewer video frames significantly reduces training time, it also exacerbates error accumulation and introduces inconsistencies in the generated videos. To address these issues, we propose a Local Optimization (Local Opt.) method, which optimizes tokens within localized windows while leveraging contextual information to reduce error propagation. Inspired by Lipschitz continuity, we propose a Representation Continuity (ReCo) strategy to improve the consistency of generated videos. ReCo utilizes continuity loss to constrain representation changes, improving model robustness and reducing error accumulation. Extensive experiments on class- and text-to-video datasets demonstrate that our approach achieves superior performance to the baseline while halving the training cost without sacrificing quality.