search

cavalleria / cavaface

face recognition training project(pytorch)
MIT License
456 stars 88 forks source link

About transfer learning #49

Open FeiMiBa opened 3 years ago

FeiMiBa commented 3 years ago

Hi, I followed your instruction to train a model with MS1M training set, and I ended up achieving 99+@LFW, 95+@CFP, 95+@AgeDB. This pretrained model also achieved 97+ on my own testing set.

Then I used this pretrained model of mine to do transfer learning on my own training set, which is quite different from MS1M/LFW/CFP/AgeDB.... When I finished the tranfer learning, the final model achieved 99+ on my own testing set, but only 80+@LFW, 70+@CFP, 70+@AgeDB.

I wonder if I made any mistakes? Is there any ways to imporve the performance on my own testing set while remaining fair enough good performance on other testing sets?

Thank you for your time.

cavalleria commented 3 years ago

you can fix some layers params and finetune the remain layers, and do not overfitting in your own training set.

FeiMiBa commented 3 years ago

@cavalleria That's exactly what I did with my model. I only trained on the last two layers, unfortunately that the results i got...