search

seasonSH / DocFace

Face recognition system for ID photos
MIT License
371 stars 124 forks source link

base model training without se #22

Closed hustzeyu closed 5 years ago

hustzeyu commented 5 years ago

Hi, I am currently training the base model without se-module. The reg_loss decreased from 1.031 to 0.065 while the am loss seems stay at 16.287. Isn't training without Se-module just by commenting out the following part: if i % 2 == 1: net = se_module(net) net = net + shortcut shortcut = net print('| shortcut') at the face_resnet.py ? Would you please help me? Thank you.

hustzeyu commented 5 years ago

I have tried to train the base model with Se-module from the beginning this morning, which is your original code without your pre-trained model, but still I have got the same result. The reg_loss has reduced dramatically from 1.068 while the (am) loss flatuated around 16.3 Is it normal? Do you have the same process?