About
I am Hyunsoo Cha, a PhD student at Seoul National University, advised by Prof. Hanbyul Joo. I am currently a Research Scientist Intern at NVIDIA AI-Mediated Reality and Interaction Research Group (AMRI) in Santa Clara, CA.
My long-term research goal is to build virtual worlds in which humans and intelligent agents seamlessly interact, with agents that perceive, represent, and meaningfully assist people in their everyday lives. Toward this vision, I develop generative models that capture the richness of human appearance, motion, and behavior, serving as a foundation for embodied agents that can truly understand and collaborate with people.
News
- Jul 2026 Starting as a Research Scientist Intern at NVIDIA AI-Mediated Reality and Interaction Research Group (AMRI), Santa Clara.
- Apr 2026 Vanast has been selected as a Highlight at CVPR 2026.
- Feb 2026 Vanast has been accepted to CVPR 2026.
- Jan 2026 Durian has been accepted to ICLR 2026.
- Feb 2025 PERSE has been accepted to CVPR 2025.
- Feb 2024 PEGASUS has been accepted to CVPR 2024.
- Sep 2022 Joined SNU Visual Computing Lab as an MS/PhD student.
- Jan 2022 Started as a Research Intern at SNU Visual Computing Lab.
Experience
Mentors: Seonwook Park, Amrita Mazumdar
Collaborators: Rajarshi Roy, Koki Nagano, Christian Jacobsen
Publications
Robot-Factored World Models via Robot Rendering
Factors action realization and robot appearance out of the world model by rendering actions as visible robot geometry, yielding a shared visual interface that transfers across viewpoints and unseen robot embodiments.
SimuScene: Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image
A compositional 3D reconstruction pipeline that puts physics in the loop of shape and layout estimation, producing a stable, simulation-ready scene from a single image.
Vanast: Virtual Try-On with Human Image Animation via Synthetic Triplet Supervision
Given a human image and one or more garment images, our method generates virtual try-on with human image animation conditioned on a pose video while preserving identity.
Durian: Dual Reference Image-Guided Portrait Animation with Attribute Transfer
Given a portrait and one or more reference images specifying target attributes, Durian generates a portrait animation with attribute transfer in a single-stage pipeline.
PERSE: Personalized 3D Generative Avatars from A Single Portrait
A method for building an animatable personalized generative avatar from a reference portrait.
PEGASUS: Personalized Generative 3D Avatars with Composable Attributes
A method for constructing a personalized generative 3D face avatar from monocular videos.