Abhishek et al. found several flaws in DermaMNIST, such as improper resizing and data leakage:
https://www.nature.com/articles/s41597-025-04382-5
Not sure if this explains the bad performance of NCALab on this dataset, but it definitely makes sense to try DermaMNIST-C instead:
https://github.com/kakumarabhishek/Corrected-Skin-Image-Datasets
Abhishek et al. found several flaws in DermaMNIST, such as improper resizing and data leakage:
https://www.nature.com/articles/s41597-025-04382-5
Not sure if this explains the bad performance of NCALab on this dataset, but it definitely makes sense to try DermaMNIST-C instead:
https://github.com/kakumarabhishek/Corrected-Skin-Image-Datasets