"I find the gaps existing methods overlook — then build the data or framework to close them."

I'm a Master's researcher at Soongsil University working across Neural Rendering, Generative AI, and Embodied AI. Rather than solving well-defined problems, I start by asking what problem no one has addressed yet — and build whatever is needed to answer it.

My research follows two consistent threads:

🔹 Gap → Data   When the right dataset doesn't exist, I build it from scratch.      CulinaryCut: first physics-aware VLA benchmark via MPM simulation (ECCV 2026)      CLP: first real-world paired lens-contamination segmentation dataset (CVPR 2026)

🔸 Gap → Design   When existing representations fall short, I redesign the framework.      IPRF: intrinsic decomposition for geometry-preserving NeRF style transfer (ACMMM WS 2025)      UDT: hierarchy-aware contrastive loss for consistent diffusion transformations (BMVC 2025)      FEEL: unpaired learning to inject emotion into text-to-motion generation (CVPR WS 2025)

My strongest trait is adaptability — each of these five projects spans a different subfield, yet all follow the same problem-first philosophy.

Year	Paper	Venue	Role
2026	🟢 CulinaryCut — Physics-aware VLA Benchmark via Material Point Method		Primary
2026	🟢 CLP — Real-World Contaminated Lens Protectors Dataset for Robust Segmentation		2nd Author
2025	🟢 FEEL — Emotion-Enriched Text-to-Motion Generation via Unpaired Learning		Co-first *
2025	🟢 UDT — Unsupervised Discovery of Transformations in Diffusion Models		Co-first *
2025	🟢 IPRF — Intrinsic-Guided Photorealistic Style Transfer for Radiance Fields		Primary

_{* equal contribution}

2024 ─────────────────────────────────── 2026
  │                                        │
IPRF       UDT      FEEL      CLP    CulinaryCut
Neural    Diffusion  Motion  Dataset    Physics
Rendering   Model    Gen.    Build.      VLA

Deep Learning & Research

Tools & Web