NolanBeaujault/3DRespiView-Trainer ? reverse-engineered prompt

Build me a small Docker based project for 3DRespiView that lets me annotate chest images in the browser, train a YOLO model on those annotations, and optionally export a mobile ready model for Android.

I want a simple setup where I can run one command to start everything, open a local web app for labeling, upload JPG images, draw the thoracoabdominal area, and export the annotations in a format the training pipeline can use. Then I want a script that takes a dataset version folder, splits the images into training and validation sets, runs the YOLO training, saves the results, and produces a best.pt file. If I pass an extra option, it should also convert the final model into a compressed TFLite file and save it in an exports folder.

Please make the whole thing work cleanly with Docker, keep the data folders organized, and give me clear README instructions for the full workflow. If you need current docs for anything, look them up online.