search

mmasana / FACIL

Framework for Analysis of Class-Incremental Learning with 12 state-of-the-art methods and 3 baselines.
https://arxiv.org/pdf/2010.15277.pdf
MIT License
512 stars 98 forks source link

Imagenet100 #26

Closed zijGao closed 1 year ago

zijGao commented 1 year ago

Dear author,

Thanks for your great work. It really helps me a lot!

I wonder if you tried to use imagenet_100 (imagenet_subset) as the dataset? I tried to use the imagenet_subset dataset but failed. It really helps me if you can provide the code about using the imagenet_subset dataset!

mmasana commented 1 year ago

Hi,

happy to hear that you like our work!

We have used the imagenet_subset dataset without problem. However, you need to have the ImageNet dataset set up for yourself, since I think it is not available as a torchvision import. To add the dataset, you can follow the instructions here. Or if you have tried that, then I would need more information on which error you get.

wangsen1312 commented 1 year ago

@mmasana Thank you for your information. Could you share where we can download the imagenet100 dataset or we can generate one from original imagenet?

mmasana commented 1 year ago

You can download the dataset from their website, and then create the dataset .txt files as explained in the link from last comment.

In the case of the Imagenet subset of 100 classes, you can either create the train.txt and test.txt only containing the images from the corresponding first 100 classes, or you can use the .txt files with all classes and then use the 'class_order' option to use only those classes that correspond to the split you want.

wangsen1312 commented 1 year ago

@mmasana Thank you for your reply, it's working now!