Source preserving method improves text based human motion editing
SuperMotion: Source-Preserving Denoising for Text-Driven Human Motion Editing
Computer Vision and Pattern Recognition
Summary
Editing human motion using text instructions can change some parts while keeping others the same, but existing methods often lose smooth movement details. The authors proposed a new technique called SuperMotion that keeps using the original motion as a reference throughout the editing process to better preserve these details. Their approach blends the predicted changes with the aligned original motion, improving accuracy and keeping natural movement dynamics. This method works without needing extra information on which parts to keep unchanged.
What this means in practice
- •For game developers: Edit character animations using text commands while preserving original motion quality and timing details for more realistic gameplay.$Commercial implications: Enables creating improved character animation editing tools that game studios can sell as part of animation pipelines.
- •For film animators: Refine motion capture data with text inputs for visual effects while maintaining original movement smoothness and dynamics.
Authors
Fa-Ting Hong, Peter Wonka
Abstract
Text-driven human motion editing aims to realize a requested change while preserving compatible source content. Existing diffusion editors rely largely on learned conditioning for preservation of the unedited part, yet their outputs can lose temporal detail as denoising proceeds. We propose the \textbf{Source-Preserving Denoising framework (SuperMotion)}, which explicitly reuses the source at each reverse step for source preservation. We first align the source motion to the output timeline and predict a preservation gate that controls reuse across frames and feature dimensions. A clean-space source anchor then utilizes the learned preservation gate to blend the predicted clean motion with the aligned source and passes the corrected estimate directly to the sampling posterior. Because the aligned source is a realized motion rather than a regression output, the anchor injects sample-level temporal detail that a reconstruction-trained denoiser tends to smooth away. To learn effective source reuse, we supervise the anchored estimate against the editing target and match its second temporal differences through a temporal high-frequency loss. These objectives require no explicit edit masks. Extensive experiments show that SuperMotion improves editing accuracy, reaching 33.20\% full-pool R@1 on MotionFix, while reducing temporal-detail attenuation and preserving motion dynamics as it realizes the requested changes. Ablations confirm that the learned preservation gate is responsible for the gain and that it reuses the source to retain the unedited content properly.