caiyu6666/MedIAnomaly ? reverse-engineered prompt

Build me a Python project that benchmarks anomaly detection on medical images, like the one described in the README. I want it to support both reconstruction based methods and SSL based methods, with a simple way to train and evaluate each method from the command line. It should work with the provided seven datasets, assuming the data is placed in the home directory as described, and it should be easy to change the data path if needed.

Please include the training and evaluation flow for the main methods listed in the README, along with a clean setup that matches Python 3.10 and PyTorch 2.1.2. I also want basic visualization outputs for reconstructions and anomaly examples, so I can see what the models are doing. If anything depends on current package versions or best practices, look up the current docs online if you need to.