lingtengqiu/yolo_nano ? reverse-engineered prompt

Build me a small PyTorch project for pedestrian detection using YOLO Nano. I want to be able to train on the COCO person class, run evaluation on a saved checkpoint, and do inference on images or video with bounding boxes drawn on top.

Please include a simple training script, a testing script, and an inference script, with config files for the model and dataset paths. It should support the training tricks mentioned in the paper, like data augmentation, fixup, warmup, and cosine learning rate decay, since those seem important for getting decent results. If needed, look up the current docs online for any YOLO Nano details that help make the implementation match the paper.

Make it easy to run from the command line, and keep the code clean and minimal. I’d also like a way to download or point to the COCO dataset, and a sample setup for person only detection so I can try it quickly.