As the original authors for StyleAvatar3D were unable to publish their code, I decided to work on my own reimplementation of their paper. This will be both a reimplementation of the training and testing pipeline, but will also have additional tooling for generating and using these avatars in practice.
Star this repo and tag along and if you'd like to help out, let me know! I've only just started so if you're seeing this, come back in two weeks hopefully, there's more completed :)
The original paper: StyleAvatar3D
- The original authors have recently announced they will be releasing their code. I will still continue on my implementation, as I plan on adding additional tooling for everyday use, and not just replicating research.
Core Features
| Feature | Percentage Completed |
|---|---|
| Pose generation for depth maps (used for control net + dataset generation) | [75%] |
| EG3D port | [0%] |
| Stable diffusion integration | [0%] |
| Controlnet integration | [0%] |
| Training loop | [0%] |
| Testing loop | [0%] |
Future Features
- Training and testing GUI
- Mixamo-compatible full-body avatars