Reproduce “Image Diffusion Models Are Good 3D Human Motion Generators” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siyuan Yang, Chong Wang, Hao Cheng, Jun Liu, Bihan Wen, Alex Kot, Hedvig Kjellstrom
The authors introduce a shared text-to-motion system that represents root-centered joint motion as topology-informed images while keeping travel in a separate stream. One image-generation model handles HumanML3D and KIT-ML using their native skeletons, with dataset conditioning. On HumanML3D, shared SiT-L improves FID from 0.089 to 0.082 and Top-1 R-Precision from 0.540 to 0.546; evidence is limited to these two datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.