Suggestion of using AU R-CNN instead of OpenFace for better AU detection accuracy.

[machanic]

We have released our new research : AU R-CNN. https://github.com/sharpstill/AU_R-CNN

It achieves state-of-the-art AU detection performance and accuracy. For example, in BP4D 3-fold testing protocol, the result is 63% F1 score.

We suggest GANimation to use AU R-CNN to detect action units on human faces instead of OpenFace for better accuracy.