search

guoqiangqi / PFLD

Implementation of PFLD A Practical Facial Landmark Detector , reference to https://arxiv.org/pdf/1902.10859.pdf
626 stars 166 forks source link

Landmark not stable in video #73

Open John1231983 opened 4 years ago

John1231983 commented 4 years ago

Thanks for sharing it. I used your code and run it for video captured from camera. The landmark are not stable although the face is not moving. Could you suggest the solution to fix it? I find the pfld app and it looks stable

guoqiangqi commented 4 years ago

Using landmark tracking algorithm such as SDM did.

John1231983 commented 4 years ago

@John1231983 : Thanks. Have you use it in your method? Do you meet the issue in video?

guoqiangqi commented 4 years ago

@John1231983 : Thanks. Have you use it in your method? Do you meet the issue in video?

No,i just trained landmark detection model ,not include tracking algorithm.These landmarks detected are also not stable while using videoes for testing.

John1231983 commented 4 years ago

@guoqiangqi : could you please check the eyes with different cases: openning/closing with wflw training dataset? I checked and it is not working with these cases although dlib works well

hegc commented 4 years ago

@John1231983 when closing the eyes, the landmarks is not very accurately. Maybe there is almost no eye-closing data in the dataset?

luckynote commented 4 years ago

@John1231983 when closing the eyes, the landmarks is not very accurately. Maybe there is almost no eye-closing data in the dataset?

I think so that. I only find 338 landmarks faces of closing eyes from WFLW. It is so slowly to be training.