Gen4P: Repurposing Video Generators for 4D Perception
October 14, 2026
Gen4P transforms pretrained video generators into feed-forward 4D perceivers that recover geometry and motion from monocular video. It uses a lightweight cross-attention decoder and continuous-time queries to achieve state-of-the-art 4D attribute prediction.
HOW THIS AFFECTS YOU
●
builderThis offers a way to repurpose existing video generation models for perception tasks.
●
researcherYou can extract 4D spatial information from video without the need for iterative denoising loops.