Vision--language models (VLMs) are increasingly aligned via Group Relative Policy Optimization (GRPO)-style training. However, relying solely on terminal outcome rewards yields sparse credit assignment in multi-step reasoning, weakening the linkage between visual evidence and intermediate steps and often causing unstable optimization and visual hallucinations. We propose Differential Feedback, which automatically constructs token/step-level supervision masks by repairing erroneous reasoning trajectories, explicitly marking the key positions that require correction. Without costly large-scale step-by-step human annotations, our method enables process-level visual alignment and can be seamlessly integrated into existing GRPO-like frameworks. Experiments on multimodal reasoning benchmarks including MMMStar and MathVista show an average 3% improvement under matched compute budgets. Our approach offers an effective, low-cost solution for accurate vision--reasoning process alignment.
Thinking with Gaze: Sequential Eye-Tracking as Visual Reasoning Supervision for Medical VLMs
School of Computing, University of Georgia(佐治亚大学计算机学院);Department of Computer Science and Engineering, University of Texas, Arlington(德克萨斯大学阿灵顿分校计算机科学与工程系);Massachusetts General Hospital, Harvard Medical School(麻省总医院哈佛医学院);Department of Biomedical Engineering, New Jersey Institute of Technology(新泽西理工学院生物医学工程系)
VLM-E2E: Enhancing End-to-End Autonomous Driving with Multimodal Driver Attention Fusion
The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));Li Auto Inc.(Li汽车公司);the School of Aeronautics and Astronautics, Xiamen University(厦门大学航空航天学院);The Hong Kong University of Science and Technology(香港科技大学)
City-VLM: Towards Multidomain Perception Scene Understanding via Multimodal Incomplete Learning
The Hong Kong University of Science and Technology (Guangzhou)(香港科学与技术大学(广州));Zhejiang University(浙江大学);Fudan University(复旦大学);Harbin Institute of Technology (Shenzhen)(哈尔滨工业大学(深圳));Terminus Technologies Co., Ltd.;Pengcheng Laboratory(鹏城实验室)
Align, Don't Blend: Orthogonal Supervision Disentanglement As the Key to Generalizable AI-Generated Image Detection
Tencent Youtu Lab;East China University of Science;Shenzhen University;Hong Kong University of Science;Department of XXX, University of YYY, Location, Country;School of ZZZ, Institute of WWW, Location, Country
The Ebb and Flow of Multimodal Focus: Scheduling Visual Relay Windows for Grounded VLM Reasoning
School of Computer Science and Technology, Tongji University(同济大学计算机科学与技术学院);School of Fashion and Textiles, The Hong Kong Polytechnic University(香港理工大学纺织及制衣学院);School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院)
ReVision: A Dataset and Baseline VLM for Privacy-Preserving Task-Oriented Visual Instruction Rewriting
School of Information(信息学院);The University of Texas at Austin(德克萨斯大学奥斯汀分校);Department of Statistics and Data Science(统计与数据科学系);Yale University(耶鲁大学);School of Computing and Augmented Intelligence(计算与增强智能学院)
Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning
Fudan University(复旦大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Shanghai Jiaotong University(上海交通大学);Nankai University(南开大学);Shanghai University(上海大学);Nanyang Technological University(南洋理工大学);Hong Kong Polytechnic University(香港理工大学);Tongji University(同济大学);Nanjing University(南京大学);University of California, Merced(加州大学梅德福分校);Chinese University of Hong Kong(香港中文大学)
Red-teaming the Multimodal Reasoning: Jailbreaking Vision-Language Models via Cross-modal Entanglement Attacks
Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所);University of Chinese Academy of Sciences(中国科学院大学);People’s Public Security University of China(中国人民公安大学)