search

TadasBaltrusaitis / OpenFace

OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
Other
6.76k stars 1.83k forks source link

CE-CLM Model for landmark detection #894

Open Zye3 opened 4 years ago

Zye3 commented 4 years ago

First of all, thank you for your contribution. I want to know which part of main_ceclm_general.txt is directly related to landmark detection. Because I want to retrain the model by adding my own data set. So I want to know which ones I have to retrain. Besides, if possible, I would like you to tell me where the model_eye part, early_term_cen_of. txt, haaralAlign. txt, validator_cnn_68.txt are trained, what are their main functions？

Best Regards

TadasBaltrusaitis commented 4 years ago

Hi,

It depends on which parts you're interested in re-training. E.g. if you are using the same set/number of landmarks you may only need to retrain the patch experts, but if you need to change the landmark set you would also need to retrain the Point Distrubution Model, and the validator (if you care about tracking in videos). I can provide you more details if I better understand your use case.

Zye3 commented 4 years ago

Hello, I'm trying to retrain with a new data set. The purpose is for landmark detection. Therefore, in the ce-clmgeneral Model, I know that Retrain the patch Experts and Point Distribution Model are needed. But I'm still not sure if I need to retrain and generate new early_term_cen_of. txt, validator_cnn_68.txt, and tris_68.txt. Why? And I don't see these files in the code you provide how to train and generate them?