This project is built to easy/end-to-end extract face coeffs and rendering.
- Extractor
- DeepFace3DRecon
- extract face box / 68 keypoints (CPU / CUDA)^[only test with single face images]
- align with
lm3d_std
and re-crop - extract 257-dimensional coefficients
- get raw face mesh with type
trimesh.Trimesh
- visualize face mesh by overlay on original video
- DeepFace3DRecon
- Renderer
- PIRender
- render video by given identity image and coefficients
- PIRender
- Utils
- read any video to 4D numpy array
- write 4D uint8 numpy array to any video
Refer to example.py.
- more extractors and renders
- more tests
- more documents
This repo mainly borrows code from sicxu/Deep3DFaceRecon_pytorch and RenYurui/PIRender