search

jhb86253817 / PIPNet

Efficient facial landmark detector
MIT License
412 stars 82 forks source link

increasing accuracy if using only eye eyebrow nose landmark #48

Open john09282922 opened 1 year ago

john09282922 commented 1 year ago

Hi, thanks for great works. I have some question, if only eye, eyebrow, nose part for training, accuracy is increasing? my concern is large pose and occlusion image make some problem like lower performance even if using resnet101. Do u have some idea? If I using multiple dataset, will it be good?

thanks

jhb86253817 commented 1 year ago

Hi, if only train with thses parts, I guess the accuracy may increase a little because the boundaies are usually more difficult. But you may need to tune the hyperparameter, especially the number of neighbors. Since you have less number of landmarks, you need to try a smaller number of neighbors, e.g., 5. With more datasets, I think it will alleviate the large pose and occlusion problem. You may try it.