Spatial Intelligence Lab

Two papers are accepted to NeurIPS 2026 (one oral)

2026.09

Two papers are accepted to NeurIPS 2026 (one oral)

Two papers from our lab have been accepted to the Conference on Neural Information Processing Systems (NeurIPS) 2026, including one oral presentation (0.36% acceptance rate).

Beyond Family Labels: A Taxonomic Ornstein-Uhlenbeck Prior for Avian 3D Shape Recovery (Oral)

Selected for an oral presentation, a distinction given to only 0.36% of all submissions. Co-first authored by our students Yunchan Jeon and SooGon Kim, with Seungwook Kim, CheolWon Lee, and Prof. Jongmin Lee. The paper introduces a taxonomic Ornstein-Uhlenbeck prior that transfers 3D shape knowledge along the phylogenetic tree, enabling accurate avian 3D shape recovery even for rarely observed species.

ProxySearch: Decoupled Inference-Time Scaling for Diffusion Models via Asymmetric Noise-Rank Transfer

By Seungwook Kim and Prof. Jongmin Lee. ProxySearch decouples inference-time scaling from the base diffusion model: a distilled proxy model searches over noise candidates, and the discovered noise ranking transfers asymmetrically to the full model, improving generation quality at a fraction of the search cost.