Reproduce “Investigating the Quality of DermaMNIST and Fitzpatrick17k Dermatological Image Datasets” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kumar Abhishek, Aditi Jain, Ghassan Hamarneh
The authors audit image quality and evaluation protocols in DermaMNIST, HAM10000, and Fitzpatrick17k, and release corrected datasets. They trace images to lesion identities, remove cross-partition overlap, identify duplicate and conflicting-label images, and standardize splits. In DermaMNIST, 1,006 of 7,470 unique lesions (13.47%) cross partitions; the audit improves evaluation, but does not quantify leakage’s effect or guarantee error-free data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.