search

alldbi / FLM

Tensorflow implementation of Fast Geometrically-Perturbed Adversarial Faces
MIT License
24 stars 5 forks source link

Can you please share the code for the training processedure? #3

Closed fbiying87 closed 3 years ago

fbiying87 commented 3 years ago

Thanks for sharing your source code. It helps a lot to get your proposed idea.

I was wondering, if you can also provide the code for the training processedure? The current version works well for CASIA-Web Images, but I was wondering, if I can train it on another database.

Thanks a lot.

Have a nice day!

Best Regards!

alldbi commented 3 years ago

Hi,

I am glad that you are interested in our work. Our attack model is independent of the training model. So, we mainly used trained deep face recognition models from available repositories. Hence, you can also use any model on repositories and use it as a target model in our algorithm.

Thanks, Ali

fbiying87 commented 3 years ago

Thx for replying so quickly.

So the current class output is 10575 due to CASIA database. If I would use LFW, how should I adapt the y label. The output is also different from 10575. Can you give me a hint?

So I should change the number of output class and retrain it to store the correct model weights based on Facenet. Do you have some sample codes on that?

Thanks for your help!

Best!

PPPPPPPeng commented 2 years ago

Thx for replying so quickly.

So the current class output is 10575 due to CASIA database. If I would use LFW, how should I adapt the y label. The output is also different from 10575. Can you give me a hint?

So I should change the number of output class and retrain it to store the correct model weights based on Facenet. Do you have some sample codes on that?

Thanks for your help!

Best!

Have you solve this problem?