Yeobin Hong

News

• Jun 2026 InverseCrafter is accepted to ECCV 2026.
• May 2026 AGSM is accepted to ICML 2026 Spotlight (top 2.2%).
• Feb 2026 FlowAlign is accepted to ICLR 2026.
• Dec 2025 Started Visiting Student (Remote) @ ETH Zürich.

Education

• KAIST Sep 2024 – Aug 2026

  M.S. in Artificial Intelligence

  Advisor: Prof. Jong Chul Ye

• Korea University Mar 2020 – Aug 2024

  B.S. in Mathematics & Artificial Intelligence

Publications

2026

• arXiv preprint
  GeoFix: Geometry Fixing for 4D Consistent Generation Project Page

  Yeobin Hong, Hyojun Go, Dohun Lee, Jeongjae Lee, Sunghwan Hong,
  Federico Tombari, Zhaochong An, Jong Chul Ye†, Konrad Schindler†, Dominik Narnhofer†

  

  GeoFix aligns videos for geometric consistency and visual quality in 4D generation using a single depth consistency reward calibrated for camera overlap, both during training and inference-time alignment.

• ECCV 2026
  InverseCrafter: Efficient Video ReCapture as a Latent Domain Inverse Problem

  Yeobin Hong*, Suhyeon Lee*, Hyungjin Chung†, Jong Chul Ye†

  

  InverseCrafter is an inpainting inverse solver that reformulates novel view video generation as an inpainting problem solved in the latent space.

• ICML 2026 Spotlight (536/6352=2.2%)
  Alignment-Guided Score Matching for Text-to-Image Alignment in Diffusion Models Project Page

  Jaa-Yeon Lee*, Yeobin Hong*, Taesung Kwon, Jong Chul Ye

  

  AGSM improves text–image alignment using score-level preference optimization with a Plackett–Luce objective without an explicit reward. By explicitly guiding positive and negative pairs within the diffusion score-matching loss, AGSM reduces SoftREPA’s failure modes while improving generation and editing across different backbones.

• ICLR 2026
  FlowAlign: Trajectory-Regularized, Inversion-Free Flow-based Image Editing Code

  Jeongsol Kim*, Yeobin Hong*, Jonghyun Park, Jong Chul Ye

  

  FlowAlign is an inversion-free, flow-based image editing method that achieves smooth and stable trajectories through a flow-matching loss. It also enables reversible editing by leveraging the intrinsic properties of ODE-based transformations.

• WACV 2026
  DreamMakeup: Face Makeup Customization using Latent Diffusion Models

  Geon Yeong Park*, Inhwa Han*, Serin Yang*, Yeobin Hong*, Seongmin Jeong, Heechan Jeon, Myeongjin Goh, Sung Won Yi, Jin Nam, Jong Chul Ye

  

  DreamMakeup is a face makeup customization method that can get various conditioning as input - reference images, RGB colors, textual descriptions.

2024

• ACM MM 2024
  GaussianTalker: Real-Time High-Fidelity Talking Head Synthesis with Audio-Driven 3D Gaussian Splatting Code

  Kyusun Cho*, Joungbin Lee*, Heeji Yoon*, Yeobin Hong, Jaehoon Ko, Sangjun Ahn, Seungryong Kim

  

  GaussianTalker is a framework for real-time generation of pose-controllable talking heads. It leverages the fast rendering capabilities of 3D Gaussian Splatting (3DGS) while addressing the challenges of directly controlling 3DGS with speech audio.

Experience

• ETH Zürich PRS Group Dec 2025 – May 2026

  Visiting Student (Remote)

  Advisors: Prof. Konrad Schindler, Dr. Dominik Narnhofer

• Korea University CVLAB Jan 2024 – Aug 2024

  Undergraduate Research Intern

  Advisor: Prof. Seungryong Kim