Get labels for full training set

[wmorning]

Hi Joshua!

Thanks for making this dataset publicly available. We think this may be a cool dataset to look at in some experiments that we and probably others are working on. To that end, we were wondering if you plan to do a similar human labeling process to get labels for the full training set. This may help in setting up CIFAR-10h for use as a standard benchmark to test model calibration and reliability.

We look forward toward experimenting with this dataset, and of course, citing your paper :)

Best,
Warren

[jcpeterson]

@wmorning, I'm glad you're finding it useful. We are indeed considering collecting labels for the training set. We also have another set of human-derived labels for the test set that will hopefully be released this year.