Shanlin Sun (孙 山林)

Into everything vision-related.

personal_page_profile.jpg

I am a research scientist at Meta, working on video generation.

Previously, I graduated as Ph.D. from the Department of Computer Science at University of California, Irvine (UCI), advised by Prof. Xiaohui Xie, focuing on 3D recontrsuction and generation. During my PhD, I was fortunate to intern at NEC Labs America with Bingbing Zhuang, Ziyu Jiang and Manmohan Chandraker, building neural world simulation.

Earlier, I obtained my Master’s degree from University of Southern California (USC) and Bachelor’s degree from Beihang University (BUAA).

My primary research interests are at the intersection of computer vision, computer graphics and generative AI. Happy to be connected at sslgive[at]gmail[dot]com.

news

Sep 08, 2025 Join Meta as Research Scientist.
Aug 12, 2025 Successfully defense my PhD thesis “Towards World Simulation: Object, Human and Scene Modeling with Neural Networks”. Committee members: Prof. Xiaohui Xie (Chair), Prof. Alexander C. Berg and Prof. Charless C. Fowlkes.
Jun 26, 2025 Our work Ouroboros on cycle-consistent rendering via diffusion models was accepted at ICCV 2025.
Jul 19, 2024 Our paper Hybrid-CSR was accepted as oral presentation at BMVC 2024.
Jun 24, 2024 After two-year revisions, our paper “Medical Image Registration via Neural Fields” is finally published in Medical Image Analysis!
Jun 23, 2024 Join NEC Labs America as Research Intern in Media Analytics team, mentored by Bingbing Zhuang.
Apr 23, 2024 Our paper “LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes” was selected as highlights of CVPR 2024 and will be oral presented at CVPR 2024 Workshop DDADS.

selected publications

  1. ICCV
    iccv25_ouroboros.png
    Ouroboros: Single-step Diffusion Models for Cycle-consistent Forward and Inverse Rendering
    Shanlin Sun*, Yifan Wang*, Hanwen Zhang*, and 5 more authors
    Oct 2025

    Finetune single-step diffusion models for cycle-consistent fast neural image decomposition and composition.

  2. arXiv
    coma.png
    CoMA: Compositional Human Motion Generation with Multi-modal Agents
    Shanlin Sun*, Gabriel Araujo*, Jiaqi Xu*, and 5 more authors
    arXiv preprint, Dec 2024

    How MLLM reasoning faciliates human motion generation, editing and self-correction?

  3. CVPR
    cvpr24_lidarf.gif
    LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes
    Shanlin Sun, Bingbing Zhuang, Ziyu Jiang, and 3 more authors
    In Computer Vision and Pattern Recognition , Jun 2024

    How to encode Lidar geometry guidance into NeRF training? How to deal with noisy/sparse Lidar supervision? How to augment data from Lidar data?

  4. MedIA
    nir.png
    Medical Image Registration via Neural Fields
    Shanlin Sun, Kun Han, Chenyu You, and 8 more authors
    Medical Image Analysis, Jul 2024

    Introducing neural fields into 3D medical image registration. It is lightweight and fast, can be displacement-based or velocity-based.

  5. CVPR
    cvpr22_ndf.png
    Topology-preserving Shape Reconstruction and Registration via Neural Diffeomorphic Flow
    Shanlin Sun, Kun Han, Deying Kong, and 3 more authors
    In Computer Vision and Patern Recognition , Jun 2022

    Introcuding topology preserving to template-based deep implicit shape representation via Neural ODE.