search

mysee1989 / TCAE

Self-supervised Representation Learning from Videos for Facial Action Unit Detection
162 stars 27 forks source link

Could you please share your pre-trained model? #11

Open ilovecv opened 4 years ago

ilovecv commented 4 years ago

Hi,

Thank you very much for sharing the codes. I am wondering if it is possible for you to share the pre-trained model? Thanks!

zcx-language commented 4 years ago

Thank you very much for sharing the codes. I am wondering if it is possible for you to share the pre-trained model? Thanks! My e-mail: D201980974@hust.edu.cn

mysee1989 commented 4 years ago

The model has been public.

zcx-language commented 4 years ago

Thanks!

ilovecv commented 4 years ago

@mysee1989 Hi, thank you very much for sharing your model. When I used the model to extract the expression feature and train a linear classifier on top of it, I can only get a mean F1 score of 50.80 on BP4D dataset. However, in your paper, you reported a mean F1 score of 56.1 in Table 1. I used the default hyperparameters shared in your model. Thanks.

mysee1989 commented 4 years ago

Please check you face detection, alignment and cropping step.
Different detection & cropping strategy will lead to bad results as facial AU detection are known to be sensitive to face cropping strategy. If you adopt the released model, you should align the facial images by Seetaface. https://github.com/seetafaceengine/SeetaFace2

havefunbb commented 3 years ago

can you share the pre-trained model of "au_model"?