mlmed/torchxrayvision

Built on PyTorch, it provides unified preprocessing pipelines and dataset loaders for multiple public chest X-ray cohorts (NIH, CheXpert, MIMIC-CXR, PadChest, RSNA) that can be swapped interchangeably for robust model evaluation across distributional shifts. Pre-trained DenseNet121 and ResNet50 classifiers output multi-label pathology predictions (18 conditions), while additional baseline models enable demographic prediction and anatomical segmentation via PSPNet, alongside representation learning through pre-trained autoencoders.

1,122 stars and 64,525 monthly downloads. No commits in the last 6 months. Available on PyPI.