Gen4P: Video Generators are 4D Perceivers
Gen4P transforms a video generator into a single-step 4D perceiver for recovering full 3D geometry and motion from monocular video. It achieves this by using a direct regression network instead of iterative denoising, allowing all four modalities to be predicted simultaneously in one forward pass. This framework enables state-of-the-art performance on tracking and depth estimation while offering better temporal flexibility and geometric accuracy than prior methods.