search

cleardusk / 3DDFA_V2

The official PyTorch implementation of Towards Fast, Accurate and Stable 3D Dense Face Alignment, ECCV 2020.
MIT License
2.88k stars 508 forks source link

How to get the BFM model without neck? #101

Open CZ-Wu opened 3 years ago

CZ-Wu commented 3 years ago

Excellent work! I have a question about how you get the BFM model without neck? I find that the training data you provided (https://drive.google.com/file/d/0B7OEHD3T4eCkVGs0TkhUWFN6N1k/view?usp=sharing) has 53215 vertices, but the data in bfm_noneck_v3.pkl has only 38365 vertices. If I train the model with the 300W_LP data you provided, can I use the .mat file of each image as GT directly? Looking for your reply, thank you!

cleardusk commented 2 years ago

Compared to the original bfm, bfm_noneck_v3 clips the shape/exp params, removes the ear/neck region. You can find the mapping across the vertices of the mean shape.