# Jiaya JIA 賈佳亞 Chair Professor ![Jiaya JIA](../portraits/jia.jpg) [官方档案](https://cse.hkust.edu.hk/admin/people/faculty/profile/jia) · [个人主页](https://jiaya.me/home) · [HTML](../../index.html#faculty-jia) ## 学术指标 | 指标 | 数值 | | --- | ---: | | [累计引用](https://openalex.org/A5052856441) | 76,881 | | h-index | 109 | | i10-index | 256 | | 2024–2026 收录论文 | 88 | | 在组成员 | 30 | | 产业合作机构 | 3 | 数据日期：2026-10-17。[指标定义与分位标准](../../index.html#standards)。 ### 网络关注 - [Wikipedia 人物条目](https://en.wikipedia.org/wiki/Jiaya_Jia)：2025-10-01 至 2026-09-30，1,117 次浏览。 ## 教育与履历 ### 教育 - 2004 · [HKUST](https://jiaya.me/home) · PhD, Computer Science ### 任职与研究经历 - 2003–2004 · [Microsoft Research](https://jiaya.me/home) · Visiting Scholar - 2004–2024；2015 晋升教授 · [CUHK](https://jiaya.me/home) · Assistant Professor; Professor - 2007 · [Adobe Research](https://cse.hkust.edu.hk/admin/people/faculty/profile/jia) · Collaborative research - 2019– · [SmartMore](https://jiaya.me/home) · Founder and Board Chairman - 2024– · [HKUST](https://jiaya.me/home) · Chair Professor [2019 年创办 SmartMore，任 Founder and Board Chairman。](https://www.bschool.cuhk.edu.hk/wp-content/uploads/gaf-2023-jiaya-jia-bio-en.pdf) ## 研究方向 ### [视觉理解与计算成像](https://cse.hkust.edu.hk/admin/people/faculty/profile/jia) 图像视频理解、恢复、分割、计算成像；具有多年视觉与图形学积累。 ### [多模态与生成模型](https://jiaya.me/publications) VLM、omni-cognition、视频生成和编辑。 ### [模型推理与 RL](https://jiaya.me/publications) 长链推理、奖励学习、视觉推理与 VLA。 2011—2012 图像平滑与结构提取获得多年后的 test-of-time 认可；2023—2024 的 LISA、LongLoRA、Mini-Gemini 将视觉研究与语言模型连接；2024—2026 的 Step-DPO、Lyra、VisionReasoner 扩展到细粒度奖励、语音与多任务感知。由公开论文可见明显扩张，多模态感知和可控生成维持连续；研究强度与个人指导参与程度需分别询问。 ### 长期视觉基础上的大模型平台与方法扩展 既有视觉算法和当前跨模态模型形成较宽研究组合；选题同时包含模型系统、训练目标与评测。 ## 实验室与成员 [Jiaya Jia 研究组](https://jiaya.me/home) HKUST CSE Clear Water Bay, Hong Kong 视觉理解与计算成像；多模态与生成模型；模型推理与 RL ### 在组成员（30） | 成员与公开档案 | 身份与时间 | 教育背景 | 研究与经历 | 量化记录 | | --- | --- | --- | --- | --- | | [LU Changsheng](https://jiaya.me/people) | Postdoc；2025 - present | | | — | | [WANG Yibo](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [YAO Huanjin](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [ZHOU Yiduan](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [GONG Yi](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [SUN Benteng](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [WANG Jian](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [PENG Shangpin](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [ZHAO Xiaorui](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [TUO Kaiwen](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [ZHENG Shenghe](https://jiaya.me/people) | PhD, CSE；2026 - present | | | — | | [HUANG Jiehui](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [ZHANG Wenhu](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [YE Jinhui](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [CHEN Yuxin](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [ZHANG Xichen](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [WANG Zixuan](https://jiaya.me/people) | PhD, CSE；2025 - present | | | — | | [CHU Meng](https://jiaya.me/people) | PhD, CSE；2024 - present | | | — | | [QU Tianyuan](https://jiaya.me/people) | Ph.D. student；2024 - present | | | — | | [LIANG Fengfeng](https://jiaya.me/people) | PhD, CSE；2024 - present | | | — | | [LIU Yuqi](https://jiaya.me/people) | Ph.D. student；2024 - present | | | — | | [ZHU Zheng](https://jiaya.me/people) | Ph.D. student；2024 - present | | | — | | [YANG Senqiao](https://jiaya.me/people) | Ph.D. student；2024 - present | | | — | | [GUI Haokun](https://jiaya.me/people) | PhD, CSE；2024 - present | | | — | | [WU Sitong](https://jiaya.me/people) | Ph.D. student；2023 - present | | | — | | [ZHU mingkang](https://jiaya.me/people) | Ph.D. student；2023 - present | | | — | | [XIA Bin](https://jiaya.me/people) | Ph.D. student；2023 - present | | | — | | [YU Shaozuo](https://jiaya.me/people) | Ph.D. student；2022 - present | | | — | | [LUO, Haotian](https://facultyprofiles.hkust.edu.hk/profiles.php?profile=jiaya-jia-jia) | PhD, CSE | | | — | | [ZHANG, Yuwei](https://facultyprofiles.hkust.edu.hk/profiles.php?profile=jiaya-jia-jia) | PhD, CSE | | | — | ### 校友与去向（61） | 成员与公开档案 | 身份与时间 | 教育背景 | 研究与经历 | 量化记录 | | --- | --- | --- | --- | --- | | [LU Fanbin](https://jiaya.me/people) | Ph.D.；2021 - 2026 | | Research Scientist @ Tencent | — | | [LI Jingyao](https://jiaya.me/people) | Ph.D.；2022 - 2026 | | Research Scientist @ Baidu | — | | [WANG Chengyao](https://jiaya.me/people) | Ph.D.；2022 - 2026 | | Research Scientist @ Tencent | — | | [PENG Bohao](https://jiaya.me/people) | Ph.D.；2022 - 2026 | | Research Scientist @ Huawei | — | | [LIU Shaoteng](https://jiaya.me/people) | Ph.D.；2021 - 2025 | | Research Scientist @ Adobe Research | — | | [ZHANG Yuechen](https://jiaya.me/people) | Ph.D.；2021 - 2025 | | Research Scientist @ Xiaomi MIMO | — | | [LI Yanwei](https://jiaya.me/people) | Ph.D.；2020 - 2024 | | Assistant Professor @ SJTU | — | | [ZHONG Zhisheng](https://jiaya.me/people) | Ph.D.；2021 - 2025 | | Researcher @ Tencent | — | | [CHEN Yixin](https://jiaya.me/people) | Ph.D.；2020 - 2024 | | Researcher | — | | [LAI Xin](https://jiaya.me/people) | Ph.D.；2020 - 2024 | | Researcher at ByteDance | — | | [CHEN Yukang](https://jiaya.me/people) | Ph.D.；2020 - 2024 | | Researcher at NVidia | — | | [CHU Ruihang](https://jiaya.me/people) | Ph.D.；2020 - 2024 | | Researcher at Alibaba | — | | [QIAN Shengju](https://jiaya.me/people) | Ph.D.；2019 - 2023 | | Researcher at Tencent | — | | [YANG Zetong](https://jiaya.me/people) | Ph.D.；2019 - 2023 | | Researcher at PJLab | — | | [HU Tao](https://jiaya.me/people) | Ph.D.；2019 - 2023 | | Researcher at ByteDance | — | | [LI Wenbo](https://jiaya.me/people) | Ph.D.；2019 - 2023 | | Researcher at Huawei | — | | [LU Liying](https://jiaya.me/people) | MPhil.；2019 - 2023 | | Ph.D. at EPFL | — | | [CHEN Yilun](https://jiaya.me/people) | Ph.D.；2018 - 2022 | | Researcher at Shanghai AI Laboratory | — | | [CUI Jiequan](https://jiaya.me/people) | Ph.D.；2018 - 2022 | | Professor at Hefei University of Technology | — | | [TIAN Zhuotao](https://jiaya.me/people) | Ph.D.；2018 - 2022 | | Assistant Professor at HIT | — | | [XU Xiaogang](https://jiaya.me/people) | Ph.D.；2018 - 2022 | | Assistant Professor at Zhejiang University | — | | [CHEN Pengguang](https://jiaya.me/people) | Ph.D.；2018 - 2022 | | Research Leader at SmartMore | — | | [QI Lu](https://jiaya.me/people) | Ph.D.；2017 - 2021 | | Professor at Wuhan University | — | | [WU Ruizheng](https://jiaya.me/people) | Ph.D.；2017 - 2021 | | Research Leader at SmartMore | — | | [JIANG Li](https://jiaya.me/people) | Ph.D.；2017 - 2021 | | Assistant Professor at CUHK (Shenzhen | — | | [LIN Huaijia](https://jiaya.me/people) | Ph.D.；2017 - 2021 | | Researcher at ZOOM | — | | [WANG Yi](https://jiaya.me/people) | Ph.D.；2017 - 2021 | | Researcher at Shanghai AI Laboratory | — | | [WANG Ruixing](https://jiaya.me/people) | Ph.D.；2016 - 2021 | | Leader at DJI | — | | [CHEN Yingcong](https://jiaya.me/people) | Ph.D.；2016 - 2020 | | Assistant Professor at HKUST (Guangzhou | — | | [GAO Hongyun](https://jiaya.me/people) | MPhil.；2015 - 2019 | | PhD student at University of Cambridge | — | | [ZHAO Hengshuang](https://jiaya.me/people) | Ph.D.；2015 - 2019 | | Assistant Professor at HKU | — | | [QI Xiaojuan](https://jiaya.me/people) | Ph.D.；2014 - 2018 | | Associate Professor at HKU | — | | [ZHANG Hong](https://jiaya.me/people) | MPhil.；2015 - 2018 | | Researcher at Baidu | — | | [ZHOU Chao](https://jiaya.me/people) | MPhil.；2014 - 2018 | | Co-founder of SmartMore | — | | [LI Ruiyu](https://jiaya.me/people) | Ph.D.；2014 - 2018 | | Co-founder of SmartMore | — | | [LIU Shu](https://jiaya.me/people) | Ph.D.；2014 - 2018 | | Co-founder of SmartMore | — | | [TAO Xin](https://jiaya.me/people) | Ph.D.；2013 - 2018 | | Researcher at Kuaishou | — | | [LIN Di](https://jiaya.me/people) | Ph.D.；2012 - 2016 | | Associate Professor at Tianjin University | — | | [SHEN Xiaoyong](https://jiaya.me/people) | Ph.D.；2013 - 2016 | | Co-founder of SmartMore | — | | [LIAO Renjie](https://jiaya.me/people) | M.Phil.；2013 - 2015 | | Assistant Professor at UBC | — | | [SHI Jianping](https://jiaya.me/people) | Ph.D.；2011 - 2015 | | Research Director at SenseTime | — | | [YAN Qiong](https://jiaya.me/people) | Ph.D.；2008 - 2013 | | Scientist at SenseTime | — | | [LU Cewu](https://jiaya.me/people) | Ph.D.；2008 - 2013 | | Shanghai Jiao Tong University Distinguished Professor, Changjiang Scholar Distinguished Professor of the Ministry of Education China | — | | [XU Li](https://jiaya.me/people) | Ph.D.；2007 - 2010 | | Chairman and CEO at SenseTime Group Ltd | — | | [ZHANG Qi](https://jiaya.me/people) | M.Phil.；2012 - 2014 | | Sr. Director, Product & Innovation at NEX Team Inc | — | | [ZHENG Shicheng](https://jiaya.me/people) | M.Phil.；2011 - 2013 | | Quantitative Researcher at Citadel Securities | — | | [WANG Liwei](https://jiaya.me/people) | M.Phil.；2011 - 2013 | | Assistant Professor at CUHK | — | | [WU Tai-Pang](https://jiaya.me/people) | Postdoc；2010 | | CTO at Jiuling | — | | [XIA Yang](https://jiaya.me/people) | Master；2011 - 2013 | | Tech Lead Manager, Staff Software Engineer at Google | — | | [DAI Zhenlong](https://jiaya.me/people) | M.Phil.；2009 - 2012 | | Portfolio Manager at WorldQuant | — | | [QIN Zenglu](https://jiaya.me/people) | M.Phil.；2008 - 2012 | | | — | | [XU Yi](https://jiaya.me/people) | M.Phil.；2009 - 2011 | | | — | | [HUNG Chun Ho](https://jiaya.me/people) | M.Phil.；2009 - 2011 | | Software Engineer at Quora | — | | [LIU Renting](https://jiaya.me/people) | M.Phil.；2006 - 2008 | | | — | | [CHEN Jianing](https://jiaya.me/people) | M.Phil.；2007 - 2009 | | | — | | [CHUNG Hin Shun](https://jiaya.me/people) | M.Phil.；2007 - 2009 | | | — | | [FENG Wei](https://jiaya.me/people) | Postdoc；2008 | | Professor at Tianjin University | — | | [XU Leilei](https://jiaya.me/people) | M.Phil.；2005 - 2007 | | | — | | [XIONG Wei](https://jiaya.me/people) | M.Phil.；2005 - 2007 | | Partner, GenAI Infrastructure & Solutions, Microsoft CoreAI | — | | [TAO Chenjun](https://jiaya.me/people) | M.Phil.；2005 - 2008 | | | — | | [SHAN Qi](https://jiaya.me/people) | M.Phil.；2006 - 2008 | | | — | ## 业界合作 Microsoft Research（research visit）；Adobe Research（research collaboration） - **[Microsoft Research](https://cse.hkust.edu.hk/admin/people/faculty/profile/jia) · 访问研究**。2003–2004 访问研究 - **[Adobe Research](https://cse.hkust.edu.hk/admin/people/faculty/profile/jia) · 研究合作**。2007 年合作研究 - **[Adobe Systems Incorporated](https://facultyprofiles.hkust.edu.hk/profiles.php?profile=jiaya-jia-jia) · 资助研究项目**。Generative AI；负责人；2025– - **[SmartMore](https://www.bschool.cuhk.edu.hk/wp-content/uploads/gaf-2023-jiaya-jia-bio-en.pdf) · 创办企业**。2019 年创办，任董事长；CUHK 2023 年公开讲者履历所载。 ## 学术贡献与服务 资深视觉研究者，现任 Chair Professor 和 Von Neumann Institute Director；计算成像的长期影响与当前大模型研究可分别核验。 - [个人主页列 ACM Fellow、IEEE Fellow。](https://jiaya.me/home) - [SIGGRAPH Asia Test-of-Time Award 2023 对应 Image Smoothing via L0 Gradient Minimization，2024 对应 Structure Extraction from Texture via Relative Total Variation；另有 CVPR 2022 Best Paper Finalist。](https://jiaya.me/home) - [自 2021 年担任 TPAMI Associate Editor-in-Chief。](https://jiaya.me/home) ## 论文分析 ### 2024 · [Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs](https://arxiv.org/abs/2406.18629) arXiv 2024 - **问题**：整答案偏好学习难以识别长数学推理链中的局部错误。 - **方法**：以推理步骤为偏好单位，构建 10K step-wise pairs，并比较自生成与外部分布数据。 - **结果**：摘要报告 Qwen2-72B-Instruct 在 MATH 70.8%、GSM8K 94.0%；超过 70B 模型以少于 500 步训练获得近 3% MATH 改善。 - **实验条件**：结果依赖题目、基座和数据生成流程。 - **研究线**：把视觉组的学习系统能力扩展到细粒度推理监督。 阅读：Abstract（完整摘要）。 ### 2025 · [Lyra: An Efficient and Speech-Centric Framework for Omni-Cognition](https://openaccess.thecvf.com/content/ICCV2025/html/Zhong_Lyra_An_Efficient_and_Speech-Centric_Framework_for_Omni-Cognition_ICCV_2025_paper.html) ICCV 2025 - **问题**：Omni 模型对语音及其与其他模态整合覆盖不足。 - **方法**：复用开放模型和 multi-modality LoRA，加入 latent regularizer、extractor；使用 1.5M 多模态样本与 12K 长语音样本。 - **结果**：作者报告多项视觉语言、视觉语音、语音语言任务的先进表现，并使用较少计算和训练数据。 - **实验条件**：不能仅凭摘要认定全面最优。 - **研究线**：从图像语言扩展到以语音为中心的全模态接口。 阅读：Abstract（完整摘要）。 ### 2026 · [VisionReasoner: Unified Reasoning-Integrated Visual Perception via Reinforcement Learning](https://arxiv.org/abs/2505.12081) ICLR 2026（教师出版页） - **问题**：检测、分割与计数任务需要共享感知与推理能力。 - **方法**：统一奖励机制与 multi-object cognitive learning，在输出前生成结构化推理过程。 Introduction 补充 7B 模型使用 7K 样本；奖励涵盖格式、非重复、multi-object IoU 和 L1，匹配结合 batch computation 与 Hungarian algorithm。 - **结果**：作者在 10 项任务上评估；相对 Qwen2.5VL，在 COCO、ReasonSeg、CountBench 分别报告 29.1%、22.1%、13.2% 的相对增益。 - **实验条件**：统一目标仍限定 detection、segmentation 和 counting 等感知；忠实性结论来自作者人评，独立复核与跨域验证未执行。 - **研究线**：连接既有感知任务和当前 reasoning RL。 阅读：Abstract（完整）；1 Introduction（完整）；5 Conclusion（完整）。 ### 2025 · [DreamOmni: Unified Image Generation and Editing](https://arxiv.org/abs/2412.17098) CVPR 2025；2024 初版预印本 - **问题**：文本生图模型与指令、拖拽等编辑任务缺少统一框架和规模化训练数据。 - **方法**：统一文本生成与多种编辑任务，以贴纸式元素合成高质量配对编辑数据，联合训练文本生图和编辑。 引言比较计算分配相近的 U-Net 与 DiT，采用高分辨率 latent 计算及残差连接；以 VLM 替代纯文本编码器，统一处理图像与语言提示。 - **结果**：摘要和结论报告合成数据管线增强文字、位置、数量、颜色与几何控制，联合训练同时支持概念理解和编辑要求。 - **研究线**：覆盖统一图像生成与编辑方向，与三维生成和多模态推理形成相邻分支。 阅读：Abstract（完整，CVPR 与 arXiv 交叉核对）；1 Introudction（Introduction，原文如此拼写；完整）；5 Conclusion（完整）。 ## 2024–2026 论文目录 ### 2026 - [Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond](https://doi.org/10.48550/ARXIV.2604.22748)。[CoRR](https://dblp.org/rec/journals/corr/abs-2604-22748) - [AutoDataBench: Can Agents Write the Data That Feeds the Self-Improvement Loop?](https://doi.org/10.48550/ARXIV.2609.35025)。[CoRR](https://dblp.org/rec/journals/corr/abs-2609-35025) - [Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models](https://doi.org/10.48550/ARXIV.2608.27550)。[CoRR](https://dblp.org/rec/journals/corr/abs-2608-27550) - [CoDMD: Copula-aware Distribution Matching Distillation for Fast Video Generation](https://doi.org/10.48550/ARXIV.2606.21982)。[CoRR](https://dblp.org/rec/journals/corr/abs-2606-21982) - [DreamOmni2: Multimodal Instruction-based Generation and Editing](https://openaccess.thecvf.com/content/CVPR2026/html/Xia_DreamOmni2_Multimodal_Instruction-based_Generation_and_Editing_CVPR_2026_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/XiaPZHLLTWWWYJ26) - [Efficient Long-Context Modeling in Diffusion Language Models via Block Approximate Sparse Attention](https://doi.org/10.48550/ARXIV.2605.19726)。[CoRR](https://dblp.org/rec/journals/corr/abs-2605-19726) - [GKDT: General Keypoint Detection Transformer](https://doi.org/10.1007/978-3-032-36846-1_11)。[ECCV (3)](https://dblp.org/rec/conf/eccv/LuCGWYYHJ26) - [Learning Active Perception for Pixel-Space Reasoning via Visual-Intent Stratified GRPO](https://doi.org/10.1007/978-3-032-36969-7_16)。[ECCV (2)](https://dblp.org/rec/conf/eccv/ZhuCYYZJ26) - [MemVLN: Episodic and Procedural Memory for Vision-and-Language Navigation](https://doi.org/10.48550/ARXIV.2607.23504)。[CoRR](https://dblp.org/rec/journals/corr/abs-2607-23504) - [MGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech](https://doi.org/10.1007/978-3-032-37324-3_31)。[ECCV (56)](https://dblp.org/rec/conf/eccv/WangZPYLGXLYJ26) - [Mini-Gemini: Mining the Potential of Multi-Modality Vision Language Models](https://doi.org/10.1109/TPAMI.2025.3637265)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LiZWZCCLJ26) - [PhysEditWorld: A Large-Scale Dataset Toward Physics-Editable World Models](https://doi.org/10.48550/ARXIV.2606.26694)。[CoRR](https://dblp.org/rec/journals/corr/abs-2606-26694) - [RePlan: Reasoning-Guided Region Planning for Complex Instruction-Based Image Editing](https://doi.org/10.1007/978-3-032-37718-0_35)。[ECCV (66)](https://dblp.org/rec/conf/eccv/QuKZTLPYYJ26) - [RoPE-Aware Bit Allocation for KV-Cache Quantization](https://doi.org/10.48550/ARXIV.2606.24033)。[CoRR](https://dblp.org/rec/journals/corr/abs-2606-24033) - [Scaf-GRPO: Scaffolded Group Relative Policy Optimization for Enhancing LLM Reasoning](https://proceedings.iclr.cc/paper_files/paper/2026/hash/d604a371bf833227ef4f0daf3c1c8135-Abstract-Conference.html)。[ICLR](https://dblp.org/rec/conf/iclr/ZhangWZTYHJ26) - [SearchGym: Bootstrapping Real-World Search Agents via Cost-Effective and High-Fidelity Environment Simulation](https://doi.org/10.18653/V1/2026.ACL-LONG.848)。[ACL (1)](https://dblp.org/rec/conf/acl/ZhangHZWYCZTJ26) - [StarVLA-α: Reducing Complexity in Vision-Language-Action Systems](https://doi.org/10.48550/ARXIV.2604.11757)。[CoRR](https://dblp.org/rec/journals/corr/abs-2604-11757) - [StreamArena: Toward Continuous, Interactive, and Long-Horizon Agentic Streaming Video Understanding](https://doi.org/10.48550/ARXIV.2608.05703)。[CoRR](https://dblp.org/rec/journals/corr/abs-2608-05703) - [TRAC: Teacher-Guided Token Reward with Adaptive Calibration for Robust Policy Optimization](https://doi.org/10.18653/V1/2026.ACL-LONG.2210)。[ACL (1)](https://dblp.org/rec/conf/acl/WuTZXZQYJ26) - [TraveLLaMA: A Multimodal Travel Assistant with Large-Scale Dataset and Structured Reasoning](https://doi.org/10.1609/AAAI.V40I5.37335)。[AAAI](https://dblp.org/rec/conf/aaai/ChuCGYWJ26) - [UnityShots: Memory-Driven Multi-Shot Audio-Video Generation with Boundary-Aware Gating](https://doi.org/10.48550/ARXIV.2606.21661)。[CoRR](https://dblp.org/rec/journals/corr/abs-2606-21661) - [UnityVideo: Unified Multi-Modal Multi-Task Learning for Enhancing World-Aware Video Generation](https://openaccess.thecvf.com/content/CVPR2026/html/Huang_UnityVideo_Unified_Multi-Modal_Multi-Task_Learning_for_Enhancing_World-Aware_Video_Generation_CVPR_2026_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/HuangZHGCXZTWJ26) - [Unlocking Token Rewards via Training-Free Reward Attribution](https://openaccess.thecvf.com/content/CVPR2026/html/Wu_Unlocking_Token_Rewards_via_Training-Free_Reward_Attribution_CVPR_2026_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/WuTXZLZQYJ26) - [VisionDirector: Vision-Language Guided Closed-Loop Refinement for Generative Image Synthesis](https://openaccess.thecvf.com/content/CVPR2026/html/Chu_VisionDirector_Vision-Language_Guided_Closed-Loop_Refinement_for_Generative_Image_Synthesis_CVPR_2026_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/ChuYCZZYGRTLJ26) - [VisionLeaf: Entropy-Guided Leaf-First Reasoning for Efficient and Accurate Think-with-Image](https://openaccess.thecvf.com/content/CVPR2026/html/Gui_VisionLeaf_Entropy-Guided_Leaf-First_Reasoning_for_Efficient_and_Accurate_Think-with-Image_CVPR_2026_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/GuiYZCWLWTJ26) - [VisionReasoner: Unified Reasoning-Integrated Visual Perception via Reinforcement Learning](https://proceedings.iclr.cc/paper_files/paper/2026/hash/97c2f0fac182353062d304d0322ae285-Abstract-Conference.html)。[ICLR](https://dblp.org/rec/conf/iclr/LiuQZPLYJ26) - [VP-VLA: Visual Prompting as an Interface for Vision-Language-Action Models](https://doi.org/10.48550/ARXIV.2603.22003)。[CoRR](https://dblp.org/rec/journals/corr/abs-2603-22003) - [WorldLine: Action-Driven Visual Simulation for Robotic Manipulation](https://doi.org/10.48550/ARXIV.2609.38059)。[CoRR](https://dblp.org/rec/journals/corr/abs-2609-38059) ### 2025 - [ARPO:End-to-End Policy Optimization for GUI Agents with Experience Replay](https://doi.org/10.48550/ARXIV.2505.16282)。[CoRR](https://dblp.org/rec/journals/corr/abs-2505-16282) - [DLoFT: Gradient-Decoupled Fine-Tuning for Generalizable Long Chain-of-Thought Reasoning](http://papers.nips.cc/paper_files/paper/2025/hash/19c9708f31ec44b5b1cbd67f91d05d95-Abstract-Conference.html)。[NeurIPS](https://dblp.org/rec/conf/nips/WuTLZQYJ25) - [Does Your Vision-Language Model Get Lost in the Long Video Sampling Dilemma?](https://doi.org/10.1109/ICCV51701.2025.01942)。[ICCV](https://dblp.org/rec/conf/iccv/QuTPYYJ25) - [DreamOmni2: Multimodal Instruction-based Editing and Generation](https://doi.org/10.48550/ARXIV.2510.06679)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-06679) - [DreamOmni3: Scribble-based Editing and Generation](https://doi.org/10.48550/ARXIV.2512.22525)。[CoRR](https://dblp.org/rec/journals/corr/abs-2512-22525) - [DreamOmni: Unified Image Generation and Editing](https://openaccess.thecvf.com/content/CVPR2025/html/Xia_DreamOmni_Unified_Image_Generation_and_Editing_CVPR_2025_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/XiaZLWWW0J25) - [DreamVE: Unified Instruction-based Image and Video Editing](https://doi.org/10.48550/ARXIV.2508.06080)。[CoRR](https://dblp.org/rec/journals/corr/abs-2508-06080) - [DynamicBench: Evaluating Real-Time Report Generation in Large Language Models](https://doi.org/10.48550/ARXIV.2506.21343)。[CoRR](https://dblp.org/rec/journals/corr/abs-2506-21343) - [Effective LLM Knowledge Learning via Model Generalization](https://doi.org/10.48550/ARXIV.2503.03705)。[CoRR](https://dblp.org/rec/journals/corr/abs-2503-03705) - [Enhancing LLM Knowledge Learning through Generalization](https://doi.org/10.18653/V1/2025.FINDINGS-EMNLP.469)。[EMNLP (Findings)](https://dblp.org/rec/conf/emnlp/ZhuCWYZJ25) - [Exploiting Discriminative Codebook Prior for Autoregressive Image Generation](https://doi.org/10.48550/ARXIV.2508.10719)。[CoRR](https://dblp.org/rec/journals/corr/abs-2508-10719) - [From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models](https://doi.org/10.48550/ARXIV.2510.05095)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-05095) - [Generative Video Propagation](https://openaccess.thecvf.com/content/CVPR2025/html/Liu_Generative_Video_Propagation_CVPR_2025_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/LiuWWLZLLYLKJ25) - [GRADEO: Towards Human-Like Evaluation for Text-to-Video Generation via Multi-Step Reasoning](https://proceedings.mlr.press/v267/mou25a.html)。[ICML](https://dblp.org/rec/conf/icml/MouXHYJ25) - [InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy](https://doi.org/10.48550/ARXIV.2510.13778)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-13778) - [LLaMA-VID: An Image is Worth 2 Tokens in Large Language Models](https://doi.org/10.1007/978-3-031-72952-2_19)。[ECCV (46)](https://dblp.org/rec/conf/eccv/LiWJ24) - [LLMGA: Multimodal Large Language Model Based Generation Assistant](https://doi.org/10.1007/978-3-031-72920-1_22)。[ECCV (38)](https://dblp.org/rec/conf/eccv/XiaWTWJ24) - [Logits-Based Finetuning](https://doi.org/10.18653/V1/2025.EMNLP-MAIN.745)。[EMNLP](https://dblp.org/rec/conf/emnlp/LiYWSZXJ25) - [LogoSticker: Inserting Logos Into Diffusion Models for Customized Generation](https://doi.org/10.1007/978-3-031-73027-6_21)。[ECCV (60)](https://dblp.org/rec/conf/eccv/ZhuCWZJ24) - [LYRA: An Efficient and Speech-Centric Framework for Omni-Cognition](https://doi.org/10.1109/ICCV51701.2025.00352)。[ICCV](https://dblp.org/rec/conf/iccv/ZhongWLYTZLQLCYWLLJ25) - [MagicMirror: ID-Preserved Video Generation in Video Diffusion Transformers](https://doi.org/10.1109/ICCV51701.2025.01342)。[ICCV](https://dblp.org/rec/conf/iccv/ZhangLXPYLJ25) - [Mind the Interference: Retaining Pre-trained Knowledge in Parameter Efficient Continual Learning of Vision-Language Models](https://doi.org/10.1007/978-3-031-72764-1_20)。[ECCV (36)](https://dblp.org/rec/conf/eccv/TangTLHZZLJ24) - [Mixture-of-Scores: Robust Image-Text Data Valuation via Three Lines of Code](https://doi.org/10.1109/ICCV51701.2025.02281)。[ICCV](https://dblp.org/rec/conf/iccv/WuTCZLYQJ25) - [Modular Customization of Diffusion Models via Blockwise-Parameterized Low-Rank Adaptation](https://doi.org/10.48550/ARXIV.2503.08575)。[CoRR](https://dblp.org/rec/journals/corr/abs-2503-08575) - [MR-GSM8K: A Meta-Reasoning Benchmark for Large Language Model Evaluation](https://openreview.net/forum?id=br4H61LOoI)。[ICLR](https://dblp.org/rec/conf/iclr/ZengC0JJ25) - [QuickLLaMA: Query-aware Inference Acceleration for Large Language Models](https://aclanthology.org/2025.coling-main.34/)。[COLING](https://dblp.org/rec/conf/coling/LiSWZLJ0J25) - [RTime-QA: A Benchmark for Atomic Temporal Event Understanding in Large Multi-modal Models](https://doi.org/10.48550/ARXIV.2505.19125)。[CoRR](https://dblp.org/rec/journals/corr/abs-2505-19125) - [Seg-Zero: Reasoning-Chain Guided Segmentation via Cognitive Reinforcement](https://doi.org/10.48550/ARXIV.2503.06520)。[CoRR](https://dblp.org/rec/journals/corr/abs-2503-06520) - [SmartSwitch: Advancing LLM Reasoning by Overcoming Underthinking via Promoting Deeper Thought Exploration](https://doi.org/10.48550/ARXIV.2510.19767)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-19767) - [STEVE: A Step Verification Pipeline for Computer-use Agent Training](https://doi.org/10.48550/ARXIV.2503.12532)。[CoRR](https://dblp.org/rec/journals/corr/abs-2503-12532) - [Stratified GRPO: Handling Structural Heterogeneity in Reinforcement Learning of LLM Search Agents](https://doi.org/10.48550/ARXIV.2510.06214)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-06214) - [TGDPO: Harnessing Token-Level Reward Guidance for Enhancing Direct Preference Optimization](https://proceedings.mlr.press/v267/zhu25c.html)。[ICML](https://dblp.org/rec/conf/icml/ZhuCW0ZJ25) - [Training-Free Efficient Video Generation via Dynamic Token Carving](http://papers.nips.cc/paper_files/paper/2025/hash/75b2a6632ab7b9196e9cb93750986e45-Abstract-Conference.html)。[NeurIPS](https://dblp.org/rec/conf/nips/ZhangXXLPTWLJ25) - [TraveLLaMA: Facilitating Multi-modal Large Language Models to Understand Urban Scenes and Provide Travel Assistance](https://doi.org/10.48550/ARXIV.2504.16505)。[CoRR](https://dblp.org/rec/journals/corr/abs-2504-16505) - [VisionReasoner: Unified Visual Perception and Reasoning via Reinforcement Learning](https://doi.org/10.48550/ARXIV.2505.12081)。[CoRR](https://dblp.org/rec/journals/corr/abs-2505-12081) - [VisionThink: Smart and Efficient Vision Language Model via Reinforcement Learning](http://papers.nips.cc/paper_files/paper/2025/hash/88be023075a5a3ff3dc3b5d26623fa22-Abstract-Conference.html)。[NeurIPS](https://dblp.org/rec/conf/nips/YangLLWLMYZJ25) - [VisionZip: Longer is Better but Not Necessary in Vision Language Models](https://openaccess.thecvf.com/content/CVPR2025/html/Yang_VisionZip_Longer_is_Better_but_Not_Necessary_in_Vision_Language_CVPR_2025_paper.html)。[CVPR](https://dblp.org/rec/conf/cvpr/YangCTWL0J25) - [ViSurf: Visual Supervised-and-Reinforcement Fine-Tuning for Large Vision-and-Language Models](https://doi.org/10.48550/ARXIV.2510.10606)。[CoRR](https://dblp.org/rec/journals/corr/abs-2510-10606) - [VLPose: Bridging the Domain Gap in Pose Estimation With Language-Vision Tuning](https://doi.org/10.1109/TPAMI.2025.3594097)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LiCJLXJ25) ### 2024 - [BAL: Balancing Diversity and Novelty for Active Learning](https://doi.org/10.1109/TPAMI.2023.3345844)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LiCYLJ24) - [ControlNeXt: Powerful and Efficient Control for Image and Video Generation](https://doi.org/10.48550/ARXIV.2408.06070)。[CoRR](https://dblp.org/rec/journals/corr/abs-2408-06070) - [Empowering Visual Creativity: A Vision-Language Assistant to Image Editing Recommendations](https://doi.org/10.48550/ARXIV.2406.00121)。[CoRR](https://dblp.org/rec/journals/corr/abs-2406-00121) - [Generalized Parametric Contrastive Learning](https://doi.org/10.1109/TPAMI.2023.3278694)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/CuiZTLYJ24) - [GroupContrast: Semantic-Aware Self-Supervised Representation Learning for 3D Understanding](https://doi.org/10.1109/CVPR52733.2024.00470)。[CVPR](https://dblp.org/rec/conf/cvpr/Wang00TPZJ24) - [LISA: Reasoning Segmentation via Large Language Model](https://doi.org/10.1109/CVPR52733.2024.00915)。[CVPR](https://dblp.org/rec/conf/cvpr/LaiTCLY0J24) - [LongLoRA: Efficient Fine-tuning of Long-Context Large Language Models](https://openreview.net/forum?id=6PmJoRfdaK)。[ICLR](https://dblp.org/rec/conf/iclr/ChenQTLL0J24) - [MOODv2: Masked Image Modeling for Out-of-Distribution Detection](https://doi.org/10.1109/TPAMI.2024.3412004)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LiCYLJ24a) - [MR-BEN: A Comprehensive Meta-Reasoning Benchmark for Large Language Models](https://doi.org/10.48550/ARXIV.2406.13975)。[CoRR](https://dblp.org/rec/journals/corr/abs-2406-13975) - [MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs](http://papers.nips.cc/paper_files/paper/2024/hash/d81cb1f4dc6e13aeb45553f80b3d6837-Abstract-Conference.html)。[NeurIPS](https://dblp.org/rec/conf/nips/ZengLWLCDYXQZSL24) - [OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation](https://doi.org/10.1109/CVPR52733.2024.02013)。[CVPR](https://dblp.org/rec/conf/cvpr/PengWJCZTJ24) - [PFENet++: Boosting Few-Shot Semantic Segmentation With the Noise-Filtered Context-Aware Prior Mask](https://doi.org/10.1109/TPAMI.2023.3329725)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LuoTZYTJ24) - [Prompt Highlighter: Interactive Control for Multi-Modal LLMs](https://doi.org/10.1109/CVPR52733.2024.01255)。[CVPR](https://dblp.org/rec/conf/cvpr/ZhangQP0J24) - [RL-GPT: Integrating Reinforcement Learning and Code-as-policy](http://papers.nips.cc/paper_files/paper/2024/hash/31f119089f702e48ecfd138c1bc82c4a-Abstract-Conference.html)。[NeurIPS](https://dblp.org/rec/conf/nips/LiuYHLC0LJ24) - [RoboCoder: Robotic Learning from Basic Skills to General Tasks with Large Language Models](https://doi.org/10.48550/ARXIV.2406.03757)。[CoRR](https://dblp.org/rec/journals/corr/abs-2406-03757) - [SaCo Loss: Sample-Wise Affinity Consistency for Vision-Language Pre-Training](https://doi.org/10.1109/CVPR52733.2024.02583)。[CVPR](https://dblp.org/rec/conf/cvpr/WuTTC0J24) - [Scalable Language Model with Generalized Continual Learning](https://openreview.net/forum?id=mz8owj4DXu)。[ICLR](https://dblp.org/rec/conf/iclr/PengT0YJ24) - [Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs](https://doi.org/10.48550/ARXIV.2406.18629)。[CoRR](https://dblp.org/rec/journals/corr/abs-2406-18629) - [TagCLIP: Improving Discrimination Ability of Zero-Shot Semantic Segmentation](https://doi.org/10.1109/TPAMI.2024.3454647)。[IEEE Trans. Pattern Anal. Mach. Intell.](https://dblp.org/rec/journals/pami/LiCQLJ24) - [Unified Language-Driven Zero-Shot Domain Adaptation](https://doi.org/10.1109/CVPR52733.2024.02209)。[CVPR](https://dblp.org/rec/conf/cvpr/YangT0J24) - [Video-P2P: Video Editing with Cross-Attention Control](https://doi.org/10.1109/CVPR52733.2024.00821)。[CVPR](https://dblp.org/rec/conf/cvpr/LiuZ00J24)