3D Human Motion Differencing

1University of Trento 2LIGM, Ecole des Ponts, IP Paris, Univ Gustave Eiffel, CNRS 3Max Planck Institute for Intelligent Systems

Qualitative results on MotionFix dataset

In this gallery, we show samples from the MotionFix dataset. For each sample, our goal is to describe the difference between the source and target motions. The bottom displays the captions generated by Llama-diff, Qwen3-VL-8B, InternVL-3.5-8B, DEMO, our MotionDelta, and GT.