Oxford's Holo4D uses holistic 4D reconstruction as geometric control for video diffusion

Oxford · hf · 2026-10-30

A paper from the University of Oxford and NTU (Yushi Lan, Zeren Jiang, Andrea Vedaldi et al.) proposes Holo4D: it first performs holistic 4D reconstruction from input video, then uses the reconstructed 3D trajectory scaffolds as geometric control signals to guide video diffusion models in synthesizing novel-view videos while preserving geometry and motion. Project page is available.

Original post →

More from Multimodal

Multimodal channel →