pjreddie/darknet ? reverse-engineered prompt

Build me a fast object detection framework like Darknet that can train and run YOLO models on both CPU and GPU.

I want it to compile cleanly from source, with a simple build setup, and include the core pieces for loading network configs, training on image data, running inference on a single image or webcam, and saving detected boxes with labels. Please make sure the common YOLO style sample configs and data files work, and that the project is easy to run from the command line without a lot of setup.

If anything in the current docs is unclear, look up the latest usage details online and match the usual Darknet workflow as closely as possible. Also include a short, clear README with the main commands for building, training, and testing so someone new can get it running quickly.