Chenxin Li

I am a Ph.D. student at The Chinese University of Hong Kong, advised by Prof. Yixuan Yuan. I received my M.Eng from Xiamen University under Prof. Xinghao Ding and Prof. Yue Huang, where I also earned my B.Eng. Recently, I am working on creative and efficient vision algorithms to perceive and manipulate the physical world.

Email  /  CV  /  Google Scholar  /  Github  /  Twitter

profile photo
Latest News

  • [09/2024] One paper about tuning SAM for uncertainty modeling accepted to NeurIPS 2024.
  • [09/2024] One paper about prompt tuning accepted to EMNLP 2024.
  • [07/2024] One paper about tuning SAM accepted to ACM MM 2024.
  • [07/2024] One paper about biomedical omni-modal learning accepted to ECCV 2024,
  • [06/2024] Three paper (Endora + EndoSpase + Lightweight GS) accepted to MICCAI 2024. EndoGaussian is pre-printed.
  • [11/2023] One paper (FS-6DPose) accepted to 3DV 2024.
  • [07/2023] Invited talk for StegaNeRF at AIxMed Seminar, Massachusetts General Hospital and Harvard Medical School.
  • [07/2023] One paper (StegaNeRF) accepted to ICCV 2023.
Main Publications
U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation
Chenxin Li*, Xinyu Liu*, Wuyang Li*, Cheng Wang*, Hengyu Liu, Yixuan Yuan (* Equal Contribution)
Arxiv, 2024
[Project] [ArXiv] [Code]

The endeavours unveil valuable insights and sheds light on the prospect that with U-KAN, you can make strong backbone for medical image segmentation and generation.

GaussianStego: A Generalizable Stenography Pipeline for Generative 3D Gaussians Splatting
Chenxin Li*, , Hengyu Liu*, Zhiwen Fan, Wuyang Li, Yifan Liu, Panwang Pan, Yixuan Yuan (* Equal Contribution)
Arxiv, 2024
[Project] [ArXiv] [Code]

An initial exploration into embedding customizable, imperceptible, and recoverable information within the renders produced by off-the-line 3D generative models, while ensuring minimal impact on the rendered content's quality.

GTP-4o: Modality-prompted Heterogeneous Graph Learning for Omni-modal Biomedical Representation
Chenxin Li, Xinyu Liu*, Cheng Wang*, Yifan Liu, Weihao Yu, Jing Shao, Yixuan Yuan (* Equal Second-author Contribution)
ECCV, 2024
[Project] [ArXiv] [Code]

An pioneering foray into the intriguing realm of embedding, relating and perceiving the heterogeneous patterns from various biomedical modalities holistically via a graph theory.

Endora: Video Generation Models as Endoscopy Simulators
Chenxin Li*, Hengyu Liu*, Yifan Liu*, Brandon Y. Feng, Wuyang Li, Xinyu Liu, Zhen Chen, Jing Shao, Yixuan Yuan (* Equal Contribution)
MICCAI, 2024
[Project] [ArXiv] [Video] [Code]

A pioneering exploration into high-fidelity medical video generation on endoscopy scenes

EndoSparse: Real-Time Sparse View Synthesis of Endoscopic Scenes using Gaussian Splatting
Chenxin Li, Brandon Y. Feng*, Yifan Liu, Hengyu Liu, Cheng Wang, Weihao Yu, Yixuan Yuan* (* Equal Advising)
MICCAI, 2024
[Project] [ArXiv] [Code]

A first exploration of sparse-view endoscopic scene reconstruction

LGS: A Light-weight 4D Gaussian Splatting for Efficient Surgical Scene Reconstruction
Hengyu Liu*, Yifan Liu*, Chenxin Li*, Wuyang Li, Yixuan Yuan (* Equal Contribution)
MICCAI, 2024
[Project] [ArXiv] [Code]

A first exploration of endoscopic scene reconstruction in an extremly storage-limited setting.

EndoGaussian: Gaussian Splatting for Deformable Surgical Scene Reconstruction
Yifan Liu*, Chenxin Li*, Chen Yang, Yixuan Yuan (* Equal Contribution)
Preprint, 2024
[Project] [ArXiv] [Video] [Code]

Real-time surgical reconstruction with Gaussian Splatting representation

StegaNeRF: Embedding Invisible Information within Neural Radiance Fields
Chenxin Li*, Brandon Y. Feng*, Zhiwen Fan*, Panwang Pan, Zhangyang Wang (* Equal Contribution)
International Conference on Computer Vision (ICCV), 2023
[Project] [ArXiv] [Video] [Code]

NeRF with multi-modal IP information instillation

Knowledge Condensation Distillation
Chenxin Li, Mingbao Lin, Zhiyuan Ding, Nie Lin, Yihong Zhuang, Xinghao Ding, Yue Huang, Liujuan Cao
European Conference on Computer Vision (ECCV), 2022
[PDF] [Supp] [ArXiv] [Code]

Co-design of dataset and model distillation


Projects

    Embedding Information within Neural Radiance Fields                     
        Fig1. Rendering Views           Fig2. Residual Error (x5).           Fig3. Residual Error (x25).     Fig4. Recovered Customized Images

    Recent advances in Neural Radiance Field (NeRF) imply a future of widespread visual data distributions through sharing NeRF model weights. In StegaNeRF, we signify an initial exploration into the novel problem of instilling customizable, imperceptible, and recoverable information to NeRF renderings, with minimal impact to rendered images. We sincerely hope this work can promote the concerns about the intellectual property of INR/NeRF.


    Efficient Knowledge Distillation Algorithms

      Fig1. Knowledge Condensation Distillation     Fig2. Relation of Condensed Knowledge.       Fig3. Hint-Dynamic Distillation.

    Knowledge distillation (KD) plays a key role in developing lightweight deep networks by transferring the dark knowledge from a high-capacity teacher network to strengthen a smaller student one. In KCD (ECCV'22), we explore an efficient knowledge distillation framework by co-designing model distillation and knowledge condensation, which dynamically identifies and summarizes the informative knowledge points as a compact knowledge set across the knowledge transfer.

    In HKD, we investigate the diverse guidance effect from the knowledge of teacher model in different instances and learning stages. The existing literature keeps the fixed learning fashion to handle these knowledge hints. In comparison, we present to leverage the merits of meta-learning to customize a specific distillation fashion for each instance adaptively and dynamically.


    Data-Efficient Learning for Medical Imaging Analysis

        Fig1. GVS for Pseudo-Healthy Synthesis           Fig2. Uncertainty-Aware Self-Training                    Fig3. Enhanced Feature by GCN-DE

    Pseudo-Healthy Synthesis: As a variant of style-transfer task, synthesizing the healthy counterpart from the lesion regions is a important problem in clinical practice. In GVS (MICCAI'21), we leverage the more accurate lesion attribution by constructing an adversarial learning framework between the pseudo-healthy generator and lesion segmentor.

    Domain Adaptation/Generalization: Generalizing the deep models trained on one data source to other datasets is essential issue in practical medical imaging analysis. We present a domain adaptive approach by leveraging the self-supervised strategy called Vessel-Mixing (ICIP'21), which is driven by the geometry characteristics of retinal vessels. We also attempt tp address the domain generalization problem in medical imaging via Task-Aug (CBM'21). We investigate the neglected issue summarized as task over-fitting, that is, the meta-learning framework gets over-fitting to the simulated meta-tasks, and present a task augmentation strategy.

    Semi-Supervised Learning: The existing semi-supervised methods mainly exploit the unlabeled data via a self-labeling strategy. In UAST (NCA'21), we present to decouple the unreliable connect between the decision boundary learning and pseudo-label evaluation. We instead leverage an uncertainty-aware self-training paradigm by modeling the accuracy of pseudo-labels via uncertainty modeling.

    Few-shot Learning: Existing few-shot segmentation methods tend to fail in the incongruous foreground regions of support and query images. We present a few-shot learning method called GCN-DE (CBM'21) which leverages a global correlation capture and discriminative embedding to address the above issue.

Professional Activities

Conference Reviewer

ICML'24, ICLR'24, NeurIPS'23-24, CVPR'23-24, ICCV'23, ACM MM'23-24, MICCAI'23-24

Journal Reviewer

TIP, DMLR, PR, TNNLS, NCA


Modified from Jon Barron