Susung Hong

space.jpg

Welcome to my website! I am a Ph.D. student in GRAIL at the University of Washington, co-advised by Ira Kemelmacher-Shlizerman, Steve Seitz, and Brian Curless.

My research focuses on introducing new control mechanisms and perception for generative models, and the orchestration of models that could outperform humans in creating sophisticated content and constructing virtual worlds.

news

Mar 18, 2025 Paper, code, and weights of our new work, MusicInfuser, have been released.
Feb 26, 2025 Our two papers, Perturb-and-Revise and Spatiotemporal Skip Guidance (STG) have been accepted to CVPR 2025.
Jan 15, 2025 I will be joining Adobe Research as a research intern this summer!
Sep 25, 2024 Our two papers, Smoothed Energy Guidance (SEG) and Effective Rank Regularizaiton for 3DGS have been accepted to NeurIPS 2024.
May 02, 2024 Our ReDream paper has been accepted to ICML 2024.
Sep 22, 2023 Our Debiased-SDS paper has been accepted to NeurIPS 2023 and has been integrated into threestudio.
Jul 14, 2023 Our Self-Attention Guidance (SAG) paper has been accepted to ICCV 2023 and is available at huggingface/diffusers :hugs::firecracker:

selected publications

  1. Preprint
    musicinfuser.gif
    MusicInfuser: Making Video Diffusion Listen and Dance
    Susung Hong, Ira Kemelmacher-Shlizerman, Brian Curless, and Steven M Seitz
    Preprint, 2025
    arXiv Website
  2. CVPR
    pnr.gif
    Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories
    Susung Hong, Johanna Karras, Ricardo Martin-Brualla, and Ira Kemelmacher-Shlizerman
    CVPR, 2025
    arXiv Website
  3. CVPR
    stg.gif
    Spatiotemporal Skip Guidance for Enhanced Video Diffusion Sampling
    Junha Hyung*, Kinam Kim*, Susung Hong, Min-Jung Kim, and Jaegul Choo
    CVPR, 2025
    arXiv Website
  4. NeurIPS
    seg.jpg
    Smoothed Energy Guidance: Guiding Diffusion Models with Reduced Energy Curvature of Attention
    Susung Hong
    NeurIPS, 2024
    arXiv Website
  5. ICML
    redream.gif
    Retrieval-Augmented Score Distillation for Text-to-3D Generation
    Junyoung Seo*, Susung Hong*, Wooseok Jang*, Inès Hyeonsu Kim, Minseop Kwak, Doyup Lee, and Seungryong Kim
    ICML, 2024
    arXiv Website
  6. ICCV
    sag.jpg
    Improving Sample Quality of Diffusion Models Using Self-Attention Guidance
    Susung Hong, Gyuseong Lee, Wooseok Jang, and Seungryong Kim
    ICCV, 2023
    arXiv Website