muhammadrahman-hehe/apple_counting ? reverse-engineered prompt

Build me a Python app that can watch a conveyor belt video and count apples as they pass a counting line, not just when they appear in the frame.

It should detect each apple individually even when they are piled up and overlapping, keep tracking them across frames, and only add to the count the moment an apple crosses the line. I want it to support two classes, normal apples and damaged apples, with each class shown in a different consistent color. Please also add a simple live overlay with the total count, count by class, how many are visible right now, average confidence, processing speed, and a small running count graph.

Make it able to load a trained model and an input video, then save an output video with the boxes, labels, and counting line drawn on top. If it helps, add a small calibration tool so I can click two points to set the counting line for my belt. Use the existing project files if they are there, and look up current docs online if you need to.