PyJulie/USSL4MIC ? reverse-engineered prompt

Build me a Python project for universal semi supervised learning on medical images, based on the USSL idea from the paper. I want to be able to train a model on labeled medical images while also using unlabeled data that may include unknown classes or even images from a different domain, and have the system handle that as part of training.

Please make it work for medical classification use cases like dermatology and ophthalmology, with a simple config driven training script and a separate evaluation script. It should include the main pieces from the paper, like unknown class detection, unknown domain detection with a VAE style module, and domain adaptation during training.

I’d also like clean model and utility code, a way to point to a config file for different datasets, and sensible outputs like saved checkpoints and evaluation results. If you need to look up current docs or best practices for any libraries, go ahead and do that.