Xiaowei Hu / Professor
I am a Professor at the School of Future Technology, South China University of Technology (SCUT). Before joining SCUT, I was a Research Scientist at the Shanghai Artificial Intelligence Laboratory. I received my Ph.D. from The Chinese University of Hong Kong (CUHK), supervised by Prof. Pheng-Ann Heng and Prof. Chi-Wing Fu, and my B.Eng. from South China University of Technology (SCUT).
My research focuses on computer vision and multimodal intelligence. My current interests include: (i) Visual perception and restoration. Restoring visual content and improving perception under adverse weather and complex illumination, with an emphasis on generalization to real-world conditions; (ii) Multimodal understanding and controllable generation. Learning across vision, language, and audio for multimodal understanding, and developing methods for fine-grained control and consistency in visual generation; and (iii) Continual adaptation and self-evolving agents. Studying how agents can autonomously generate tasks and training data, learn from interaction and feedback, and sustain improvement as tasks and environments change.
Selected Publications
[1] MedZERO, NeurIPS 2026. Self-evolving agents for medical reasoning.
[2] Unveiling Deep Shadows, IJCV 2026. A shadow survey and benchmark.
[3] Real-World Weather Restoration, NeurIPS 2025. Restoration with dual-level reinforcement learning.
[4] Demystify Transformers & Convolutions, IEEE TPAMI 2025. Controlled comparisons of visual architectures.
[5] EchoTraffic, CVPR 2025. Audio-visual traffic anomaly understanding.
[6] MagicTailor, IJCAI 2025. Component-controllable image personalization.
[7] WResVLM, ECCV 2024. Vision-language-guided weather restoration.
I also contributed to InternImage, a large-scale vision foundation model based on deformable convolutions (CVPR 2023, Highlight).
Openings
We are recruiting self-motivated Postdocs, Ph.D. and Master’s students, Research Assistants, and Interns in these directions. Please send your CV and transcripts to huxiaowei@scut.edu.cn. SCUT students interested in research are also welcome to get in touch.
我们正在招募博士后、博士生、硕士生、科研助理和实习生，研究方向包括视觉感知与复原、多模态理解与可控生成、持续适应与自进化智能体。欢迎热爱科研、具备自我驱动能力的同学将简历和成绩单发送至 huxiaowei@scut.edu.cn。也欢迎有意参与科研的华工在校生联系交流。
News
[09/2026] One paper accepted to NeurIPS 2026: MedZERO, on self-evolving medical reasoning agents. Congrats to Xilin!
[03/2026] Unveiling Deep Shadows, our shadow survey and benchmark, published in IJCV. [Report]
[11/2025] Three papers accepted to AAAI 2026: IdentityStory on visual storytelling, SurgPub-Video on surgical video understanding, and GMAI-VL on medical vision-language modeling.
[09/2025] Included in the World’s Top 2% Scientists list by Stanford University for 2022–2025.
[09/2025] Two papers accepted to NeurIPS 2025: reinforcement learning for weather restoration and SceneDecorator for scene-consistent storytelling. Congrats to Fuyang and Donghao!
Earlier news
[06/2025] One paper accepted to ICCV 2025: Fast Image Super-Resolution via Consistency Rectified Flow. Congrats to Jiaqi!
[04/2025] Two papers accepted: physics-informed weather restoration to IEEE TCSVT, and MagicTailor for controllable personalization to IJCAI 2025.
[02/2025] One paper accepted to CVPR 2025: EchoTraffic, on audio-visual traffic anomaly understanding.
[12/2024] One paper accepted to IEEE TPAMI: Demystify Transformers & Convolutions.
[09/2024] One paper accepted to IEEE TIP: Video Instance Shadow Detection Under the Sun and Sky.
[07/2024] One paper accepted to ECCV 2024: WResVLM, on vision-language-guided weather restoration.
[12/2023] 河北广播电视台、河北科学技术馆联合报道：胡枭玮：科学研究·兴趣是最好的老师
Our Team
Academic Genealogy
