Which ImageNet-100?

[YugeTen]

Hi all,

First of all, thank you so much for creating this library, I have found it to be super useful for my own research!

I was wondering if you could provide some details on the ImageNet-100 dataset that you used? I cannot seem to find any "standard" ImageNet-100 dataset for downloading on the internet and the papers that use this dataset (eg. 1 and 2) seem to randomly select 100 classes from the dataset.

Any of your help would be much appreciated!

[vturrisi]

Hey,

Thanks for the comments! We are glad that this is useful for other researchers :)
About imagenet-100, you can get the list of classes from here.
We will probably add this list to our repo somewhere to avoid confusion.

[ankitpatnala]

Hi all,

First of all, thank you so much for creating this library, I have found it to be super useful for my own research!

I was wondering if you could provide some details on the ImageNet-100 dataset that you used? I cannot seem to find any "standard" ImageNet-100 dataset for downloading on the internet and the papers that use this dataset (eg. 1 and 2) seem to randomly select 100 classes from the dataset.

Any of your help would be much appreciated!

This repo is very useful tiny image-net to download images

[vturrisi]

thanks, I'll check it out

[YugeTen]

Hi @vturrisi, thanks for the swift response! Apologies, I should have done a more thorough search of the closed issues!

One more question: do you use all the images in these 100 classes?

@ankitpatnala thanks for sharing the repo :)

[vturrisi]

@YugeTen no worries! And yes, we use all the images, both for the self-supervised pertaining and the linear classifier.

[YugeTen]

Thank you!

[chenshuang-zhang]

Hey,

Thanks for the comments! We are glad that this is useful for other researchers :) About imagenet-100, you can get the list of classes from here. We will probably add this list to our repo somewhere to avoid confusion.

@vturrisi Hi! Thank you for your update of pretrained checkpoints in the model zoo recently! I hope to use the checkpoint on ImageNet-100 for some evaluations, and I have downloaded the dataset as you said from here. However, I still have a question:

What's the order of class in the final 100-dim predictions? I mean, with the pretained checkpoint in the model zoo, what class is each dimension in the final 100-dim prediction logit?

Thank you very much!

[vturrisi]

Hey @YingYingFight
Just sort the labels and you will get the correct order, e.g., labels = sorted(Path(entry.name) for entry in os.scandir(data_path) if entry.is_dir()).

[chenshuang-zhang]

Hey @YingYingFight Just sort the labels and you will get the correct order, e.g., labels = sorted(Path(entry.name) for entry in os.scandir(data_path) if entry.is_dir()).

@vturrisi Thank you for your reply!

I wonder the entry.name is like n02869837 or the class name like dog?

Thank you!

[vturrisi]

it's the former (n02869837).

[chenshuang-zhang]

it's the former (n02869837).

@vturrisi

Got it! Thank you very much for your help!

[vturrisi]

No worries! Thanks for using the library :)