deep-motion-editing

Why we included this project

Game studios, VFX houses, and animation tooling teams that work with motion-capture data will find this a self-contained reference for two hard 3D animation problems. It packages the methods from two SIGGRAPH 2020 papers: skeleton-aware motion retargeting, which reuses a captured animation on a character with a different skeleton, and unpaired motion style transfer, which can drive 3D characters directly from 2D video. The whole pipeline runs in PyTorch, from reading BVH and FBX files through training and running the networks to rendering the final result in Blender. Pretrained models and demo scripts are included, so you can reproduce the published outputs or feed in your own Mixamo and mocap data. It works as a reference implementation you can study when you need research-grade detail rather than a packaged product.