The rapid success of Vision Large Language Models (VLLMs) often depends on the high-resolution images with abundant visual tokens, which hinders training and deployment efficiency. Current training-free visual token compression methods exhibit serious performance degradation in tasks involving high-resolution, text-oriented image understanding and reasoning. In this paper, we propose an efficient visual token compression framework for text-oriented VLLMs in high-resolution scenarios. In particular, we employ a light-weight self-distillation pre-training stage to compress the visual tokens, requiring a limited numbers of image-text pairs and minimal learnable parameters. Afterwards, to mitigate potential performance degradation of token-compressed models, we construct a high-quality post-train stage. To validate the effectiveness of our method, we apply it to an advanced VLLMs, InternVL2. Experimental results show that our approach significantly reduces computational overhead while outperforming the baselines across a range of text-oriented benchmarks. We will release the models and code soon.
Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods
Institute of Trustworthy Embodied AI, Fudan University（复旦大学可信具身人工智能研究所）；Shanghai Key Laboratory of Multimodal Embodied AI, Fudan University（复旦大学上海市多模态具身人工智能重点实验室）；WeChat Vision, Tencent Inc.（腾讯微信视觉团队）；South China University of Technology（华南理工大学）
Embracing Dialectic Intersubjectivity: Coordination of Different Perspectives in Content Analysis with LLM Persona Simulation
Department of Media and Information（媒体与信息系）；Michigan State University（密歇根州立大学）；College of Liberal Arts（人文学院）；Colorado State University（科罗拉多州立大学）；Department of Communication（传播系）；Department of Advertising and Public Relations（广告与公共关系系）；Department of Communication Studies（传播学系）；Texas Christian University（德克萨斯 Christian 大学）；Departments of Communication and Political Science（传播与政治学系）；University of California, Los Angeles（加州大学洛杉矶分校）
CityRiSE: Reasoning Urban Socio-Economic Status in Large Vision-Language Models via Reinforcement Learning
Information Hub, The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州）信息中心）；Department of Electronic Engineering, BNRist, Tsinghua University（清华大学电子工程系）
Quantitative Outcome-Oriented Assessment of Microsurgical Anastomosis
Learning Algorithms and Systems Laboratory (LASA), École Polytechnique Fédérale de Lausanne (EPFL), Switzerland.（学习算法与系统实验室（LASA），瑞士联邦理工学院（EPFL））；Iuliu Hatieganu University of Medicine and Pharmacy, Cluj-Napoca, Romania.（伊利亚·哈蒂亚努大学医学与药学学院，罗马尼亚克鲁日-纳波卡）；Department of Neurosurgery, National Hospital of Denmark, Copenhagen, Denmark.（神经外科部门，丹麦国家医院，哥本哈根，丹麦）
PathReasoner-R1: Instilling Structured Reasoning into Pathology Vision-Language Model via Knowledge-Guided Policy Optimization
Harbin Institute of Technology (Shenzhen)（哈尔滨工业大学（深圳））；Microsoft Research（微软研究院）；National University of Singapore（新加坡国立大学）；Technical University of Dresden（德累斯顿技术大学）
ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining
School of Information Science and Electronic Engineering, Shanghai Jiao Tong University（上海交通大学信息科学与电子工程学院）；MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University（上海交通大学人工智能MOE重点实验室）；College of Artificial Intelligence, Nanjing Agricultural University（南京农业大学人工智能学院）；Key Laboratory of Livestock Farming Equipment, Ministry of Agriculture and Rural Affairs, Nanjing Agricultural University（南京农业大学农业机械化关键实验室）
National Taiwan University（台湾大学）；MOST Joint Research Center for AI Technology and All Vista Healthcare（科技部人工智能技术与全视角医疗联合研究中心）；Academia Sinica（中央研究院）；Umbo Computer Vision（Umbo计算机视觉公司）；Asus Intelligent Cloud Services（华硕智能云服务）
Person re-identification (re-ID) aims at matching images of the same identity across camera views. Due to varying distances between cameras and persons of interest, resolution mismatch can be expected, which would degrade person re-ID performance in real-world scenarios. To overcome this problem, we propose a novel generative adversarial network to address cross-resolution person re-ID, allowing query images with varying resolutions. By advancing adversarial learning techniques, our proposed model learns resolution-invariant image representations while being able to recover the missing details…
National Taiwan University（台湾大学）；University of Toronto（多伦多大学）；National Chiao Tung University（台湾交通大学）
Person re-identification (re-ID) aims at matching images of the same person across camera views. Due to varying distances between cameras and persons of interest, resolution mismatch can be expected, which would degrade re-ID performance in real-world scenarios. To overcome this problem, we propose a novel generative adversarial network to address cross-resolution person re-ID, allowing query images with varying resolutions. By advancing adversarial learning techniques, our proposed model learns resolution-invariant image representations while being able to recover the missing details in low-r…
