search

DonkeyShot21 / cassle

Official repository for the paper "Self-Supervised Models are Continual Learners" (CVPR 2022)
MIT License
117 stars 18 forks source link

About the Forward Transfer #11

Closed moonlitt closed 2 years ago

moonlitt commented 2 years ago

Hi, Thanks for your excellent work! I'm curious about how to calculate the "Forward Transfer" after training. For example, I have successfully re-produced the class-il results for Fine-tuning and CaSSLe (with BYOL) on Cifar-100 but don't know how to directly check the FT results. Does it need a seperate run to obtain the "linear evaluation accuracy of a random network" as the paper stated? BTW, just to be sure, is it right to directly check the "val_acc1" results of wandb board as the final linear evaluation accuracy?

DonkeyShot21 commented 2 years ago

Hi. Thanks for the interest in our work.

Yes, if I remember correctly I ran linear evaluation on a random network.

Yes, it is fine to use the val_acc1 for cifar, while for in100 and DomainNet you should run linear evaluation (for all intermediate tasks as well).

moonlitt commented 2 years ago

Got it, thanks again!