reefscape_ai

This project is the data files and labels to train a Yolov8n model.

Steps

1. Use labelme to label new images
2. Run labelme2yolo.py to convert the Labelme labels to YOLO labels python3 labelme2yolo.py --category_mapping labels.json images dataset
3. Setup Ultralytics in a virtualenv:

virtualenv -p python3 venv
pip install ultralytics

4. Retrain Yolov8n: yolo detect train data=/home/nlbutts/projects/reefscape_data/reefscape.yaml model=yolov8n.pt epochs=100 imgsz=640

5. Test it on a video clip yolo detect predict model=runs/detect/train/weights/best.pt source=../reefscape_data/images/PXL_20250107_222204416.TS.mp4

6. Export to IMX500: yolo export format=imx model=runs/detect/train5/weights/best.pt

This will export directory called best_imx_model in the runs/detect/trainX/weights directory.

SCP this folder to a Pi and then run TBD to generate a packed model.