ieee8023/covid-chestxray-dataset

The dataset includes hierarchical multi-label annotations covering 19 pathology conditions beyond COVID-19 (bacterial, fungal, viral pneumonias, ARDS, MERS, SARS), with curated severity scores, lung segmentations, and bounding boxes contributed by multiple research groups. Images are sourced from publications, hospitals, and public radiology platforms, integrated with a PyTorch dataloader for direct ML pipeline compatibility and available in medical formats (DICOM, NIfTI). Notably, the project emphasizes research ethics—approved by institutional review boards—and explicitly discourages unvalidated diagnostic claims, distinguishing it from competitive benchmarking datasets.

3,055 stars. No commits in the last 6 months.