urchin-detection

Purple sea urchin detection for Apsis clinic project

YOLOv9

Setup steps

1. Clone this repository
2. Create and activate a Python environment using venv python3 -m venv venv source venv/bin/activate
3. Download all images and labels from google drive. Put in a download folder with images and labels subfolders
4. Clone the YOLOv9 repo using git clone https://github.com/WongKinYiu/yolov9.git

Steps for training the YOLOv9 model

1. First, activate your virtual environment using the command source venv/bin/activate, replacing venv with the name of your virtual environment
2. Run data.py. This will take your input images and labels and organize them into the proper folders for YOLO training and evaluation. If you've run data.py before, your previous data folder will be deleted and replaced with a the new one. Run python3 data.py -h to see information about what flags you can pass to run the model on different datasets.
3. To run the YOLOv9 model: yolo detect train model=[ABSOLUTE PATH TO FILE 'yolov9c.pt] data = [ABSOLUTE PATH TO FILE 'urchins.yaml'] epochs = 50 batch = 2. Note that batch = 3 may also work, but batch = 4 crashes the clinic computer (but may work on a more computationally efficient system).

YOLOv8

Setup steps

1. Clone this repository
2. Create and activate a Python environment using venv python3 -m venv venv source venv/bin/activate
3. Download all images and labels from google drive. Put in a download folder with images and labels subfolders
4. Clone the YOLOv8 repo using git clone https://github.com/ultralytics/ultralytics.git
5. Install requirements for yOLOv8 pip install ultralytics

Steps for training the YOLOv8 model

1. First, activate your virtual environment using the command source venv/bin/activate, replacing venv with the name of your virtual environment
2. Run data.py. This will take your input images and labels and organize them into the proper folders for YOLO training and evaluation. If you've run data.py before, your previous data folder will be deleted and replaced with a the new one. Run python3 data.py -h to see information about what flags you can pass to run the model on different datasets.
3. To run the YOLOv8 model: yolo detect train data='[INSERT ABSOLUTE PATH TO YOUR 'urchins.yaml' FILE HERE]' model=yolov8n.pt epochs=50 batch=4 freeze=10

Steps for running model inference on an image, video, or stream

1. To get predictions for a particular image or video, run yolo predict model='[PATH TO TRAINED MODEL .pt FILE]' source='[PATH TO IMAGE, PATH TO VIDEO, OR STREAM ID TO RUN INFERENCE ON]. Can use show parameter to see inference being done in real time.
2. To run object tracking on a particular video, run yolo track model='[PATH TO TRAINED MODEL .pt FILE]' source='[PATH TO IMAGE, PATH TO VIDEO, OR STREAM ID TO RUN TRACKING ON]. Can also set conf=[SOME FLOAT], iou=[SOME FLOAT] and the show parameter allows you to see the video inference being done in real time.

Example running command for robot video stream yolo track model='/Users/fwright/Library/CloudStorage/GoogleDrive-fwright@g.hmc.edu/My Drive/clinic/urchin-detection/runs/detect/train26/weights/best.pt' source='rtsp://192.168.2.2:8554/video_udp_stream_0' show

YOLOv5

Setup steps

1. Clone this repository
2. Create and activate a Python environment using venv python3 -m venv venv source venv/bin/activate
3. Download all images and labels from google drive. Put in a download folder with images and labels subfolders
4. Clone the YOLOv5 repo and install requirements git clone https://github.com/ultralytics/yolov5 pip install -U -r yolov5/requirements.txt

Steps for training the YOLOv5 model

1. First, activate your virtual environment using the command source venv/bin/activate, replacing venv with the name of your virtual environment
2. Run data.py. This will take your input images and labels and organize them into the proper folders for YOLO training and evaluation. If you've run data.py before, your previous data folder will be deleted and replaced with a the new one. Run python3 data.py -h to see information about what flags you can pass to run the model on different datasets.
3. To run the YOLOv5 model: python3 yolov5/train.py --data urchins.yaml --weights yolov5s.pt --epochs 50 --batch 4 --freeze 10

References

YOLOv5 transfer learning based on tutorial: https://kikaben.com/yolov5-transfer-learning-dogs-cats/#yolov5-transfer-learning-execution Object tracking from: https://docs.ultralytics.com/modes/track/