lifewatch/planktonclass ? reverse-engineered prompt

Build me a local Python tool for classifying plankton images from folders of labeled classes.

I want to point it at a project folder with images arranged in class subfolders, then be able to create the project, validate the config, train a model, and generate a report after training. It should also let me start a local browser based API and UI for testing predictions, and copy a set of notebooks into the project for exploring the data, augmentation, prediction results, and explainability.

Please include support for downloading a pretrained model, and make it easy to run a quick demo project out of the box. I’d also like a way to package a trained model into a Docker image for inference. If you need to check current docs or best practices for the ML and notebook setup, look them up online first.