About the use of free model about class-incremental-learning

[yaoyao-liu]

Hi @kkmtmkk,

Thanks for your interest in our work. free_model is a model interface for our other project. In this work, it is set as None as follows,
https://github.com/yaoyao-liu/mnemonics-training/blob/42890b9c468ca73334354f29a0d1fd16ed9b9453/1_train/trainer/mnemonics.py#L100
We don't use it during training. So you may directly ignore this variable.

If you have any further questions, feel free to leave additional comments here or send me an email.

Best,
Yaoyao

[kangmintong]

Thanks for your clear explanations.
Also, I have another question. When I want to reproduce your result in table1 which reaches 64.95 on cifar100, I find the hyperparameters set in source code is inconsistent with that in your essay. I'll appreciate it if you can give some clues on the parameters when reproducing the result 64.95 on cifar100.
Thanks.
Best,
Mintong

[yaoyao-liu]

Hi @kkmtmkk,

We fix some bugs in the previous code and update the results in the latest arXiv version. Please refer to these results.

To reproduce the latest results, you need to change the baseline to LUCIR, i.e., replacing https://github.com/yaoyao-liu/mnemonics-training/blob/master/1_train/trainer/incremental.py with https://github.com/yaoyao-liu/class-incremental-learning/blob/main/trainer/incremental_lucir.py. I'll merge the two projects later.

Best,
Yaoyao

[kangmintong]

Thanks for your explanation.

Best,
Mintong