About Me

I am currently a Postdoctoral Research Fellow in MBZUAI and CMU, working with Prof. Kun Zhang . I am broadly interested in computer vision and machine learning. My current research focuses on causality and representation learning in ML, and video understanding, person re-identification and trajectory prediction in CV.

Prior to that, I received my Ph.D degree at Tsinghua University, advised by Prof. Jie Zhou and Prof. Jiwen Lu . In 2016, I obtained my B.Eng. in the Department of Automation, Tsinghua University.

If you are interested in our work and want to join us, please do not hesitate to drop me an email, guangyichen1994(at)gmail.com , with your resume.

News

  • 2022-05: 1 paper on domain adaptation is accepted by ICML'2022.
  • 2022-04: 1 paper is accepted by TIP.
  • 2022-03: 1 paper is accepted by TIP.
  • 2022-03: 3 paper is accepted by CVPR'2022, 1 oral.
  • 2021-11: Our python package for causal discovery causal-learn is released. Any feedback is welcome.
  • 2021-11: I give a talk at EPFL to introduce using causal inference for computer vision task [slides], thanks Yuejiang for the invitation!
  • 2021-07: 1 paper on person re-identification and attention learning is accepted by TIP.
  • 2021-07: 3 paper on trajectory prediction and attention learning is accepted by ICCV'2021.
  • Publications

    dise Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion
    Tianpei Gu*, Guangyi Chen*, Junlong Li, Chunze Lin, Yongming Rao, Jie Zhou, and Jiwen Lu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022
    [Arxiv] [Code]

    We propose a new framework to formulate the trajectory prediction task as a reverse process of motion indeterminacy diffusion, in which we progressively discard indeterminacy from all the walkable areas until reaching the desired trajectory.

    dise DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting
    Yongming Rao, Wenliang Zhao, Guangyi Chen, Yansong Tang, Zheng Zhu, Guan Huang, Jie Zhou, and Jiwen Lu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022
    [Arxiv] [Code] [Project]

    DenseCLIP is a new framework for dense prediction by implicitly and explicitly leveraging the pre-trained knowledge from CLIP.

    dise FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessment
    Jinglin Xu*, Yongming Rao*, Xumin Yu, Guangyi Chen, Jie Zhou, and Jiwen Lu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, (Oral).
    [PDF] [Code] [Project]

    We propose a counterfactual analysis method for human trajectory prediction to alleviate the negative effects brought by environment bias.

    dise Unintentional Action Localization via Counterfactual Examples
    Jinglin Xu*, Guangyi Chen*, Jiwen Lu, and Jie Zhou
    IEEE Transactions on Image Processing (TIP), 2022
    [PDF]

    We propose an approach to disentangle the effects of content and intention clues by building a counterfactual video pool, which mitigates the negative effect brought by biased action content and highlights the causal effect of intention on model prediction.

    dise Probabilistic Temporal Modeling for Unintentional Action Localization
    Jinglin Xu*, Guangyi Chen*, Jiwen Lu, and Jie Zhou
    IEEE Transactions on Image Processing (TIP), 2022
    [PDF]

    We propose a probabilistic framework for unintentional action localization, in which we model the uncertainty of annotations with temporal label aggregation and use it for training a dense probabilistic localization model.

    dise Human Trajectory Prediction via Counterfactual Analysis
    Guangyi Chen, Junlong Li, Jiwen Lu, and Jie Zhou
    IEEE International Conference on Computer Vision (ICCV), 2021
    [Arxiv] [Code]

    We propose a counterfactual analysis method for human trajectory prediction to alleviate the negative effects brought by environment bias.

    dise Personalized Trajectory Prediction via Distribution Discrimination
    Guangyi Chen, Junlong Li, Nuoxing Zhou, Liangliang Ren, and Jiwen Lu
    IEEE International Conference on Computer Vision (ICCV), 2021
    [Arxiv] [Code]

    We present a distribution discrimination (DisDis) method to predict personalized motion patterns by distinguishing the potential distributions in a self-supervised manner.

    dise Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification
    Yongming Rao*, Guangyi Chen*, Jiwen Lu and Jie Zhou
    IEEE International Conference on Computer Vision (ICCV), 2021
    [Arxiv] [Code]

    We propose to learn the attention with counterfactual causality, which provides a tool to measure the attention quality and a powerful supervisory signal to guide the learning process.

    dise Person Re-identification via Attention Pyramid
    Guangyi Chen, Tianpei Gu, Jiwen Lu, Jin-An Bao, and Jie Zhou
    IEEE Transactions on Image Processing (TIP), 2021
    [PDF] [Supp] [Code]

    We propose attention pyramid networks by the "split-attend-merge-stack" principle to jointly learn the attentions under different scales and obtain superior performance on many person re-identification datasets.

    dise Temporal Coherence or Temporal Motion: Which is More Critical for Video-based Person Re-identification?
    Guangyi Chen*, Yongming Rao*, Jiwen Lu and Jie Zhou
    Proceedings of the European Conference on Computer Vision (ECCV), 2020
    [PDF]

    We show temporal coherence plays a more critical role than temporal motion for video-based person ReID and develop an adversarial feature augmentation to highlight temporal coherence.

    dise Deep Credible Metric Learning for Unsupervised Domain Adaptation Person Re-identification
    Guangyi Chen, Yuhao Lu, Jiwen Lu and Jie Zhou
    16th European Conference on Computer Vision (ECCV), 2020
    [PDF]

    We propose to adaptively and progressively mine credible training samples to avoid the damage from the noise of predicted pseudo labels for unsupervised domain adaptation person ReID.

    dise Deep Meta Metric Learning
    Guangyi Chen, Tianren Zhang, Jiwen Lu and Jie Zhou
    IEEE International Conference on Computer Vision (ICCV), 2019
    [PDF] [Code]

    We propose to understand the deep metric learning via meta-learning.

    dise Self-Critical Attention Learning for Person Re-Identification
    Guangyi Chen, Chunze Lin, Liangliang Ren, Jiwen Lu and Jie Zhou
    IEEE International Conference on Computer Vision (ICCV), 2019
    [PDF]

    We present a self-critical attention learning method which applies a critic module to examine and supervise the attention model.

    dise Learning Recurrent 3D Attention for Video-Based Person Re-identification
    Guangyi Chen, Jiwen Lu, Ming Yang, and Jie Zhou
    IEEE Transactions on Image Processing (TIP), 2020
    [PDF]

    We propose to recurrently discover the 3D attention regions and use the reinforcement learning for optimization.

    dise Spatial-Temporal Attention-aware Learning for Video-based Person Re-identification
    Guangyi Chen, Jiwen Lu, Ming Yang, and Jie Zhou
    IEEE Transactions on Image Processing (TIP), 2019
    [PDF]

    We propose a spatial-temporal attention to jointly discover the salient clues in both spatial and temporal domain.

    Teaching

  • TA Numerical Analysis and Algorithm, Tsinghua University.
  • TA Analog Electronic Technology Foundation, Tsinghua University.
  • TA Probabilistic and Statistical Inference (ML703), MBZUAI.
  • Academic Activities

    Competitions

  • 2nd place in Semi-Supervised Recognition Challenge at FGVC7 (CVPR 2020)
  • Academic Services

  • Co-organizer: for the ICME 2019 workshop: The Third Workshop on Human Identification in Multimedia (HIM'19) [website]
  • Conference Reviewer / Program Committee Member: CVPR, ICCV, ECCV, ICML, NeurIPS, ICLR and so on.
  • Journal Reviewer: TIP, TMM, TCSVT and so on.