FL4N: Feed-forward Latent Model for 4D Novel View Synthesis
Preprint · In submission
Hello, I'm Soowon Son
Ph.D. Student, KAIST AI
I am a Ph.D. student at KAIST advised by Seungryong Kim, and I collaborate with Junhwa Hur at Google DeepMind. I am interested in improving and leveraging the geometric properties of video models. Recently, I have been exploring video models for hand-object interaction (HOI) generation and reconstruction, and how to bring these capabilities to robotics.
News
- Nov 2026Our work TrackCraft3R has been accepted to NeurIPS 2026. 🎉
- Nov 2026New preprint FL4N and DreamX are released.
- Jun 2026Our work TETO has been accepted to ECCV 2026. 🎉
- May 2026New preprint TrackCraft3R released — repurposing video diffusion transformers for dense 3D tracking.
- Feb 2026Our work Pose-dIVE has been accepted to CVPR 2026 (Findings). 🎉
- Dec 2025New preprint DiTracker released — probing and leveraging video diffusion transformer features for robust point tracking.
- Sep 2025Our work Emergent Temporal Correspondences (DiffTrack) has been accepted to NeurIPS 2025. 🎉
- Feb 2025Our work Visual Persona has been accepted to CVPR 2025. 🎉
Experience
LG ElectronicsOct 2026 – Present
Research Intern | HOI (Hand-Object Interaction) Generation and Reconstruction
Kakao Corp.Mar 2026 – May 2026
Research Intern | Token compression in video VLM
Publications
Click any thumbnail to enlarge. · * denotes equal contribution · † denotes corresponding author.
2026
DreamX: Video Diffusion Model is a Universal Latent Action Learner
Preprint · In submission
2025
Probing and Leveraging Video Diffusion Transformer Features for Robust Point Tracking
Preprint · In submission
