Document the interface for SonadorAI data loaders, describe how to train a model
As envisioned, SonadorAI is supposed to be the bridge that allows for the rapid development of AI models created form data in Sonador. It should combine best practices (such as storing ML label data in DICOM-SR) with classes that allow for different models to be created without needing to modify code.
Document should describe:
- How data should be structured prior to attempting to train a model
- What are the inputs for a PyTorch dataset, what methods need to be modified for a particular dataset to ensure a model can be created.
- Provide an example of a simple model (such as the Chest X-Ray dataset) which implements the methods and interface