Question about start training with CIFAR-100 or ImageNet (32x32)

[headscott]

I am new to all that AI training. And I want to train your model with

make experiments/RaResNet50/.done_train

First, I got access to ImageNet. But which of all of these datasets are supposed to be used for the training? Is it "ImageNet Large-scale Visual Recognition Challenge (ILSVRC) from 2012"? and then these ones:

[Training images (Task 1 & 2)](https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train.tar). 138GB

 [Training images (Task 3)](https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train_t3.tar). 728MB

 [Validation images (all tasks)](https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar). 6.3GB.

?

These are pretty big files, so I wanted to use train(32x32) and val(32x32) or CIFAR-100 for my first try. But I can't find how to correctly setup these. If I use a script, where it puts alle the .jpg train files into /imagenet/train/{label}/{file}.jpg the command from your Makefile says "No such file or directory: '/imagenet/160/train'. If I use CIFAR-100 I didn't even find out where to get the val data for validation. I also tried to use the resize.py script from the 4th step on https://github.com/locuslab/fast_adversarial/tree/master/ImageNet which should not be needed according to your "Getting Started" step 1, because I don't want to run fast aversarial training.

So what should I do? What am I doing wrong?