MotionCrafter: Dense Geometry and Motion Reconstruction with a 4D VAE

TL;DR

MotionCrafter introduces a novel framework using a 4D VAE to jointly reconstruct 4D geometry and dense motion from monocular videos, outperforming previous methods without post-optimization.

Contribution

It proposes a new joint representation and training strategy for 4D VAE that improves geometry and motion reconstruction quality from monocular videos.

Findings

Achieves 38.64% improvement in geometry reconstruction

Achieves 25.0% improvement in motion reconstruction

Outperforms prior methods on multiple datasets

Abstract

We present MotionCrafter, a framework that leverages video generators to jointly reconstruct 4D geometry and estimate dense motion from a monocular video. The key idea is a joint representation of dense 3D point maps and 3D scene flows in a shared coordinate system, together with a 4D VAE tailored to learn this representation effectively. Unlike prior work that strictly aligns 3D values and latents with RGB VAE latents-despite their fundamentally different distributions-we show that such alignment is unnecessary and can hurt performance. Instead, we propose a new data normalization and VAE training strategy that better transfers diffusion priors and greatly improves reconstruction quality. Extensive experiments on multiple datasets show that MotionCrafter achieves state-of-the-art performance in both geometry reconstruction and dense scene flow estimation, delivering 38.64% and 25.0% improvements in geometry and motion reconstruction, respectively, all without any post-optimization. Project page: https://ruijiezhu94.github.io/MotionCrafter_Page