repeat experiment on online disjoint-CIL setup

[ray0nv3rt]

Hi, I have been trying to repeat the experiment on online disjoint-CIL setup (table 3 in the article). But even with data augmentation, the last accuracy I get using all the three seeds are around 40%. The change that I did to the experiment.sh is changing the EXP from blurry10 to disjoint. Am I miss some setting that cause this?

[hwany-j]

Hi. First of all, thank you for your interest in our paper. I double-checked our method on online disjoint-CIL setup (table3 in the article) after you posted the issue, but the result is as same as table 3. I also changed the EXP from blurry10 to disjoint in experiment.sh. Hence, to avoid something different from our repo, I recommend you that you re-clone our repository and change only EXP = disjoint in experiment.sh. Thank you for your interest and I wish our repo will help your research 🙏