Paper accepted to ICCV 2023

Abstract

A paper is accepted to ICCV 2023. TPDM greatly improves 3D voxel generative modeling from 2D diffusion priors. Take home message? Modeling the 3D prior distribution as the product of independent 2D sliced distributions work extremely well.

Hyungjin Chung, Ph.D.

Research Scientist

Generative models, Inverse problems, Multimodality, Motion, and more.