ImageNet21K data preparation for pre-training

[mxjecho]

Hi,
I want to try imagenet21k pretraining. Could you please tell me how to process the raw data of imagenet21k?

[TouvronHugo]

Hi @mxjecho,
Thanks for your message.
You mean which data augmentation used?
Best,
Hugo

[mxjecho]

No, I want to know if I need to do pre-processing on the dataset. I use the winter21 imagenet 21k dataset.

[TouvronHugo]

We didn't do any particular preprocessing, we re-use the version we have internally of the dataset which should corresponds to the official version of the ImageNet dataset

[mxjecho]

Get it. BTW, what is the 'num_class' of Imagenet 21k?

[TouvronHugo]

21844 if I'm not wrong