It seems that its well known that the MNIST handwritten digit dataset contains quite a few examples where the labels are clearly wrong and correspond to the wrong digit, some examples of people looking into this below:



I've tried to find a 'corrected' label list that fixes this, but haven't been able to.

It feels like this is something that probably exists, could anyone point me in the right direction?

  • 1
    $\begingroup$ I wasn't aware of these mistakes in the dataset. Very interesting discovery... This might be disapointing but you probably have to clear it yourself using a tool like cleanlab :/ And maybe be the first one to share your corrected version. $\endgroup$
    – Ubikuity
    Commented Oct 3, 2023 at 13:33


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.