wkentaro/labelme ? reverse-engineered prompt

Build me a desktop image annotation app like Labelme.

I want to open images and mark objects with polygons, rectangles, circles, lines, and points, then save the annotations as JSON. It should also support simple image classification flags, video annotation, and exporting datasets for common computer vision workflows like semantic segmentation, instance segmentation, and COCO style output. The app should feel like a normal GUI tool, with a clean canvas, label list, zoom and pan, undo and redo, keyboard shortcuts, and an easy way to switch labels.

If possible, include a couple of AI assisted helpers too, like turning a clicked point into a mask or polygon, and text to annotation support. Make it work on Windows, macOS, and Linux, and keep it friendly for non technical users. Look up current docs online if you need to, but keep the setup simple and the app easy to run from the command line.