---
license: mit
---
They are tensorflow shapes and labels. 
(None, 32, 32, 3), (None, 1)
Training is 100k images
Verification is 10k images.

You can check and verify the integrity if you want, and I'd advise it.
They are a direct copy of tiny imagenet 200 classes, the training and validation images prepared and separated already.

I've found better sources of them, so I wouldn't advise using this. It is available to all though.