search

deepinsight / insightface

State-of-the-art 2D and 3D Face Analysis Project
https://insightface.ai
22.75k stars 5.33k forks source link

About transfer learning #1247

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.

nttstar commented 3 years ago

You can do L2 distance loss mimic.

saadnajib commented 3 years ago

FeimiBa can i know how did you perform transfer learning ? I tried training on my own custom dataset using cosfacer50 pretrained model, after setting resume=true but on testing accuracy decreased on my custom data test set. can you guide your method to transfer learning. Thank you for your time.

sriram-dsl commented 1 day ago

@FeiMiBa can i know how did you perform tranfer learning