Gen4P: Video Generators are 4D Perceivers
Gen4P recovers 3D geometry and motion from monocular video without iterative denoising.
The framework processes video latents in a single encoder pass, then uses a lightweight decoder to predict requested attributes. Its queries allow flexible sampling across space and continuous time. A frozen video-depth adapter strengthens geometry predictions without adding trainable parameters. The authors report state-of-the-art or competitive results on 3D point tracking, video depth and camera pose estimation.
HF Daily Papers' note
The framework processes video latents in a single encoder pass, then uses a lightweight decoder to predict requested attributes. Its queries allow flexible sampling across space and continuous time. A frozen video-depth adapter strengthens geometry predictions without adding trainable parameters. The authors report state-of-the-art or competitive results on 3D point tracking, video depth and camera pose estimation.
HF Daily Papers' note
score 5