jsbroks/coco-annotator — reverse-engineered prompt

Build me a web based image annotation app for creating machine learning datasets in COCO format. I want users to log in, create or import image datasets, open images in a clean labeling workspace, and draw annotations with polygons, free form curves, masks, bounding areas, and keypoints.

The tool should let someone label one object with multiple labels, mark disconnected visible parts as the same object, track object instances, and add custom metadata to objects or annotations. It should also have helpful selection tools like magic wand and optional model assisted annotation, so labeling is faster. Users need to review their datasets, use API endpoints to analyze the data, import existing COCO annotations, and export everything back to COCO format.

Please make it feel like a practical annotation tool rather than a demo. Use a Vue style frontend with a Python backend and database like the original project suggests. If you need exact COCO format details, look up the current docs online.