Can you provide the link on how the provided dataset was generated starting from the cifar gz file , Which normalize was used ? was any augmentation applied.
Is the validation set corrupted or clean same as the original cifar dataset?
If same as original then is it shuffled because we the labels dont match for both.
You can refer to https://github.com/izmailovpavel/neurips_bdl_starter_kit/blob/main/prepare_data.ipynb to see how the data was prepared.
Posted by: sharadmv @ Aug. 2, 2021, 8:42 p.m.