YogeshGajula/YOLO-NAS-Eye-Detection — reverse-engineered prompt

Build me a simple Python eye detection project as a clean Jupyter notebook. I want to use a dataset of face or eye images from Kaggle, bring it into Roboflow, then use a YOLO NAS style object detection model to find eyes in images.

Please make the notebook easy to run from a fresh folder. It should install what it needs, download or connect to the Roboflow dataset with an API key placeholder, load the model, run a prediction on a sample image like Human.jpg, and save an output image like prediction.jpg with boxes around the eyes. Show the before and after images in the notebook so it’s obvious that it worked.

Also include optional training and evaluation steps, plus simple training graphs for loss or model performance. Keep the code readable with short comments because I’m not very technical. If the current YOLO or Roboflow commands have changed, look up the latest docs online and use the working version.