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

How to use pretrained model on wflw for 300w #74

Open John1231983 opened 4 years ago

John1231983 commented 4 years ago

The paper mentioned that the method first trains on wflw dataset and use the weights as pretrain model and train on 300w

However, 300w just has 68 points without angle. How the method work with the 300w dataset. Does the method ignore the angle loss when training on 300w dataset?

ZhenyuYangGithub commented 4 years ago

The paper mentioned that the method first trains on wflw dataset and use the weights as pretrain model and train on 300w

However, 300w just has 68 points without angle. How the method work with the 300w dataset. Does the method ignore the angle loss when training on 300w dataset?

I have the same question here. Have you figured how to transfer learning the pre-trained model with the 300w dataset?

github-luffy commented 4 years ago

The paper mentioned that the method first trains on wflw dataset and use the weights as pretrain model and train on 300w

However, 300w just has 68 points without angle. How the method work with the 300w dataset. Does the method ignore the angle loss when training on 300w dataset?

You can try PFLD_68points_Pytorch or PFLD-68points-tensorflow