ultralytics/yolov3 ? reverse-engineered prompt

Build me a clean Python project for real time object detection using YOLOv3, with support for the standard YOLOv3 model plus the smaller and SPP versions. I want to be able to train on my own dataset, validate the results, and run detection on images, videos, webcams, folders, and streams, then save the annotated outputs in a sensible place.

Please also include an easy way to export the model for deployment, since I may want to use it in different formats later. Make the default setup work smoothly from the command line, and add simple examples so I can get started quickly without digging through a lot of code. If anything depends on current library behavior, look up the latest docs online if needed.

Keep it practical and well organized, and make sure the project is ready for someone to clone, install, train, test, and run right away.