Reproduce “DiT-Garment: Garment Dynamics with Diffusion Transformers” 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
Antoine Dumoulin, Laurence Boissieux, Joao Regateiro, Pierre Hellier, Stefanie Wuhrer
DiT-Garment generates garment deformations across body poses without a separate model per design. It represents geometry as a UV-space position map and conditions a diffusion transformer on garment placement, body shape, pose history, and material. On synthetic tests with held-out designs, motions, and shapes, it reports 5.20 cm average vertex error, lower than DPF and ContourCraft, but not on every metric.
Key insights
- The 3D position map provides a common body-aligned geometric reference while preserving design-specific mesh structure, seams, folds, and wrinkles.
- Modeling direct posing probabilistically is appropriate when the observed body motion does not uniquely determine the hidden cloth state; diffusion can represent multiple plausible deformations.
- Smaller UV patches improve local detail and generalization but increase transformer cost, while collision guidance improves contact validity at a substantial inference-speed penalty.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.