search

cleardusk / 3DDFA_V2

The official PyTorch implementation of Towards Fast, Accurate and Stable 3D Dense Face Alignment, ECCV 2020.
MIT License
2.9k stars 514 forks source link

Facial Landmarks #69

Open Elsaam2y opened 3 years ago

Elsaam2y commented 3 years ago

Thanks for this great work. 1- I am trying to train using more expressions and I wanted to ask about the facial landmarks, how did you get them? I have used Dlib detector but it failed for detecting high posed images, so I got only the 3dmm parameters for 300k images, not the whole dataset (I am extracting the 3dmm parameters based on HPEN paper). I have also tried to use (http://rasbt.github.io/mlxtend/user_guide/image/extract_face_landmarks/) which might be similar to 3ddfa, but still the final testing results are just grey images. 2- Is there any limitation on the size of the dataset? should using only 300k images be fine! 3- Have you used train_aug_120x120 for training? 4- Should using 199, 29 parameters work much better than 40,10?

sorry for my many questions and thanks again for this amazing work.

cyjouc commented 3 years ago

Hi,would you like share the code for training the model?Look forward to your reply! TKS

Elsaam2y commented 3 years ago

@cyjouc The training code is shared in the first version repo of 3ddfa. There you will find training folder and the train.configs which you can use to train for 12,40,10 parameters. https://github.com/cleardusk/3DDFA

cyjouc commented 3 years ago

Hi,thanks for your reply,Do the first version training code need modify?

cyjouc commented 3 years ago

Can the code be used directly? Need additional modifications?

cyjouc commented 3 years ago

Hi, There are some changes in the first version and the second version of the training. Could you please share the training process of the second version?

Elsaam2y commented 3 years ago

I am still working on it but not done yet. I will let you know if I am done with it

lhyfst commented 3 years ago

@Elsaam2y Have you done the modification of the training code?

Lego514 commented 2 years ago

Hi everyone, could you share your training code plz? I have no idea how to train the V2 version. @lhyfst @Elsaam2y @cyjouc