profile image

Pengxu Wei (魏朋旭)

HCP Lab, Sun Yat-sen University

About Me

I am an associate professor at the at the School of Computer Science and Engineering, Sun Yat-sen University (SYSU), where I work in HCP Lab (Human-Cyber-Physical Intelligence Integration Lab) supervised by Prof. Liang Lin. Before that, I received the Ph.D in Computer Science from University of the Chinese Academy of Sciences, advised by Prof. Jianbin Jiao and Prof. Qixiang Ye. I also received the Guangdong Provincial First Prize for Scientific and Technological Progress. I have been teaching Data Structures and Algorithms to undergraduate students since 2024, and Generative Artificial Intelligence to graduate students since 2025.

Research interest: Driven by the paradigm shift in Artificial General Intelligence, my research aims to bridge cognitive world modeling with generative intelligence, exploring World Models & Self-Evolving Learning to achieve open-world autonomous adaptation for realistic, complex tasks, e.g., embodied AI. To achieve this, I am also devoted to Multimodal Unified Visual Representations as the foundational features, Controllable Visual Content Generation & AIGC Detection to ensure generative trustworthiness, and Real-world Image Restoration to handle diverse, realistic situations. Together, these construct a unified closed loop for a transferable, robust, and reliable AI system.

  • 1) World Models & Self-Evolving Learning .
  • 2) Multimodal Unified Visual Representation Learning .
  • 3) Controllable Visual Content Generation & AIGC Detection .
  • 4) Real-World Image Restoration: Real-World Super-Resolution, Restoration-Recognition Alignment .

    • News

    • [Jun 2026] One CVPR paper and two ICML papers are accepted.
    • [Dec 2025] One ICCV, IJCV and two ICLR papers are accepted.
    • [Jul 2024] One paper is accepted for ECCV 2024.
    • [May 2024] Two papers are accepted for ICML 2024.
    • [Oct 2023] One paper was selected as Best Paper candidate for PRCV 2023.
    • [Jun 2023] Two papers are accepted for ICCV 2023.
    • [Apr 2023] We have released a toolbox (HCP-Diffusion) for Stable Diffusion models based on Diffusers. It facilitates flexible configurations and component support for training, in comparison with webui and sd-scripts.
    • [Mar 2023] Three papers are accepted for CVPR 2023.
    • [Feb 2023] One paper on real-world SR is accepted for TIP 2023.
    • [Feb 2023] One paper on scene graph to image synthesis is accepted for AAAI 2023.
    • [Jul 2022] One paper on adversarially-robust object detector is accepted as ECCV 2022 oral paper (2.7% of submitted papers). [Project]
    • [Mar 2022] One paper on real-world SR is accepted as CVPR 2022 oral paper (4.2% of submitted papers). [Project]
    • [Aug 2021] IEEE CASS Seasonal School on New Trends of Visual and Language Understanding was held online on August 20-22, 2021.
    • [Aug 2020] AIM 2020 Real Image Super-Resolution Challenge starts on May 8, 2020 and ends on July 17, 2021. Congratulations to the Challenge Winners! [Awards]
    • [Aug 2020] A large-scale diverse real-world image super-resolution dataset, DRealSR, is released. [Link]

    Selected Publications

    project name

    Kepler Codebook

    Junrong Lian, Ziyue Dong, Pengxu Wei*, Wei Ke, Chang Liu, Qixiang Ye, Xiangyang Ji, Liang Lin

    International Conference on Machine Learning (ICML) 2024 [Code, Paper&Supplementary]

    project name

    Image Restoration Through Generalized Ornstein-Uhlenbeck Bridge

    Conghan Yue, Zhengwei Peng, Junlong Ma, Shiyan Du, Pengxu Wei, Dongyu Zhang

    International Conference on Machine Learning (ICML) 2024 [Code, Paper&Supplementary]

    project name

    Towards Real-World Burst Image Super-Resolution: Benchmark and Method

    Pengxu Wei, Yujing Sun, Xingbei Guo, Chang Liu, Guanbin Li, Jie Chen, Xiangyang Ji, Liang Lin

    ICCV 2023 [Project, Dataset&Code, Paper]

    project name

    Masked Images Are Counterfactual Samples for Robust Fine-tuning

    Yao Xiao, Ziyi Tang, Pengxu Wei*, Cong Liu, Liang Lin

    CVPR 2023 [Project, Code, Paper]

    project name

    Taylor Neural Network for Real-World Image Super-Resolution

    Pengxu Wei, Ziwei Xie, Guanbin Li, Liang Lin*

    TIP 2023 [Code (coming soon), Paper]

    project name

    Scene Graph to Image Synthesis via Knowledge Consensus

    Yang Wu, Pengxu Wei*, Liang Lin

    AAAI 2023 [Code (coming soon), Paper]

    project name

    Adversarially-Aware Robust Object Detector

    Ziyi Dong, Pengxu Wei*, Liang Lin

    ECCV 2022 (oral) [Project, Code, Paper Supplementary]

    project name

    Dual Adversarial Adaptation for Cross-Device Real-World Image Super-Resolution

    Xiaoqian Xu, Pengxu Wei*, Weikai Chen, Yang Liu, Liang Lin and Guanbin Li

    CVPR 2022 (oral) [Project, Code, Paper]

    project name

    Cross-Domain Action Recognition via Prototypical Graph Alignment

    Binbin Yang, Junhao Zhong, Pengxu Wei*, Dongyu Zhang and Liang Lin

    ICME 2022 (oral) [Paper]

    project name

    Robust Real-World Image Super-Resolution against Adversarial Attacks

    Jiutao Yue, Haofeng Li, Pengxu Wei*, Guanbin Li and Liang Lin

    ACM Multimedia, 2021. [Paper, BibTex]

    project name

    Component Divide-and-Conquer for Real-World Image Super-Resolution

    Pengxu Wei, Ziwei Xie, Hannan Lu, Zongyuan Zhan, Qixiang Ye, Wangmeng Zuo and Liang Lin

    ECCV, 2020. [Paper, Project, Code, BibTex]

    project name

    AIM 2020 Challenge on Real Image Super-Resolution: Methods and Results

    Pengxu Wei*, Hannan Lu, Radu Timofte, Liang Lin, Wangmeng Zuo, et al.

    ECCV workshop, 2020. [Paper, BibTex]

    project name

    Deep CockTail Networks: A Universal Framework for Visual Multi-source Domain Adaptation

    Ziliang Chen, Pengxu Wei*, Jingyu Zhuang, Guanbin Li and Liang Lin

    IJCV, 2021. [Paper, BibTex]

    project name

    Deductive Reinforcement Learning for Visual Autonomous Urban Driving Navigation

    Changxin Huang, Ronghui Zhang, Meizi Ouyang, Pengxu Wei*, Junfan Lin, Jiang Su and Liang Lin

    IEEE TNNLS, 2021. [Paper, BibTex]

    project name

    Deductive Learning for Weakly-Supervised 3D Human Pose Estimation via Uncalibrated Cameras

    Xipeng Chen, Pengxu Wei* and Liang Lin

    AAAI, 2021. [Paper, BibTex]

    project name

    Min-Entropy Latent Model for Weakly Supervised Object Detection

    Fang Wan, Pengxu Wei, Zhenjun Han, Jianbin Jiao and Qixiang Ye

    CVPR 2018 [Paper]; IEEE T-PAMI, 2019 [Paper, Project&Code]

    project name

    3D Human Pose Machines with Self-Supervised Learning

    Keze Wang, Liang Lin, Chenhan Jiang, Chen Qian, Pengxu Wei

    IEEE T-PAMI 2020 [Paper] BibTex]

    project name

    Graph-Convolved Factorization Machines for Personalized Recommendation

    Yongsen Zheng, Pengxu Wei*, Ziliang Chen, Yang Cao and Liang Lin

    IEEE Transactions on Knowledge and Data Engineering (TKDE), 2021. [Paper, BibTex]

    Selected Projects

    A Near-Optimal Generative Learning Scheme for Neural Energy-Based Models

    Yang Wu, Xu cai, Pengxu Wei, Liang Lin.

    Real-World Image Super-Resolution Challenge (AIM 2020 in conjunction with ECCV 2020) [AIM2020 webpage]

    We provide high/low resolution image pairs of 3 scales captured by 5 different DSLR cameras through zooming lens. The challenging contains 3 tracks, one track per scale. The detail of the dataset is provided in Table 1. For the convergence of training, images in the trainset are cropped into fixed size patches using sliding window after careful alignment for image pairs. [Chanllenge Page]

    Organizers: Pengxu Wei (Sun Yat-Sen University), Hannan Lu (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology), Radu Timofte (ETH Zurich)

    IEEE CASS Seasonal School on New Trends of Visual and Language Understanding (NT-VLU) [Website, Video]

    To organize the IEEE CASS Seasonal School on NT-VLU, we invited several top-tier researchers to present their works and arrange two panels to discuss the new trends of NT-VLU. This seasonal school aimed to promote the frontier of language and vision research and to provide an ambiance for fruitful exchange of ideas and discussion of advances, challenges, and trends of this area in the near future. [Website]

    Spearkers: Jiebo Luo (University of Rochester), Cees G.M. Snoek (University of Amsterdam), Wen-Huang Cheng (National Yang Ming Chiao Tung University), Hanwang Zhang (Nanyang Technological University), Michael Kampffmeyer (UiT the Arctic University of Norway), Dan Xu (Hong Kong University of Sciences and Technology), Pengfei Liu (Carnegie Mellon University)

    Organizers: Liang Lin(Sun Yat-Sen University), Wangmeng Zuo (Harbin Institute of Technology), Xiaodan Liang(Sun Yat-Sen University), Yunchao Wei (Beijing Jiaotong University), Pengxu Wei (Sun Yat-Sen University), Guanbin Li(Sun Yat-Sen University)