zeyad-mansour/lunar — reverse-engineered prompt

Build me a Python project for a neural network aim assist demo for an FPS training setup. I want it to watch the screen in real time, detect player shaped targets with a YOLOv5 style PyTorch model, and use an Nvidia GPU with CUDA when available so it runs fast.

It should have one main script I can run, plus a setup mode where I can adjust sensitivity and aim movement settings without editing code. I also want a data collection mode that saves images so I can label them later and improve the detector.

Keep it simple to install with a requirements file and clear run commands. Make the code organized, with the detection, screen capture, and mouse movement logic separated so I can tweak it later for different FPS style games. If you need current PyTorch or YOLO docs, look them up online.