arthurdouillard / incremental_learning.pytorch

A collection of incremental learning paper implementations including PODNet (ECCV20) and Ghost (CVPR-W21).
MIT License
388 stars 60 forks source link

A small bug & option file for UCIR model is missing. #20

Closed suanrong closed 4 years ago

suanrong commented 4 years ago

Nice work!

A small bug I found: In your models/ucir.py:105 , test_loader should be data_loader.

I want to run experiments on ucir model. Cound you please provide an option file about ucir model?

Thanks~

arthurdouillard commented 4 years ago

Hello,

Thanks for catching this mistake, I've fixed it.

I've just pushed an option file for UCIR (https://github.com/arthurdouillard/incremental_learning.pytorch/blob/master/options/ucir/ucir_cifar100.yaml). You should use it with a data option files (see options/data).

Last time I've checked (several months ago) my UCIR was a bit worse (~2pts) than the paper UCIR, mainly because of the regularization loss. That is why in our paper PODNet, we used instead the original UCIR code base (https://github.com/hshustc/CVPR19_Incremental_Learning).

suanrong commented 4 years ago

Thanks for you reply~