GitHub repos for 3D Face Model

[shaoanlu]

3D face model can be incorporated as a post-processing solution of current faceswap model for sub-optimal performance on profile face as well as intense facial expressions. For example, matching input/output face expression.

1. eos: A lightweight header-only 3D Morphable Face Model fitting library in modern C++11/14.

• Link: https://github.com/patrikhuber/eos
• Description: EOS provides python binding and can be easily installed through pip. It supports Basel Face Model 2017 (BFM is used in Face2Face paper).

2. ExpNet: Landmark-Free, Deep, 3D Facial Expressions (Feb. 2018, First Release)

• Link: ExpNet
• Description: Pretrained weights are provided. It uses tensorflow thus can be easily ported to keras. The limitation of facial expression of their previous FacePoseNet is solved.
• Comment: ExpNet can be used as a feature extractor for perceptual loss. But it is "less able to handle extreme facial expressions" according to ExpNet paper, fig. 7.