croppie_coffee_ug

Croppie cherry detection model © 2024 by Alliance Bioversity & CIAT, Producers Direct and M-Omulimisa is licensed under GNU-GPLv3

Funded by: Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) Fair Forward Initiative - AI for All

General description

Ultralytics' Yolo V8 medium model fined tuned for coffee cherry detection using the Croppie coffee dataset.

This algorithm provides automated cherry count from RGB pictures. Takes as input a picture and returns the cherry count by class.

The predicted numerical classes correspond to the following cherry types:

{0: "dark_brown_cherry", 1: "green_cherry", 2: "red_cherry", 3: "yellow_cherry"}

Examples of use:

• yield estimates
• ripeness detection

Limitations: This algorithm does not include correction of cherry occlusion.

Note: the low visibility/unsure class was not used for model fine tuning

Repository structure

.
├── images
│   ├── foo.bar # images for the documentation
├── model_v3_202402021.pt  # fine tuning of Yolo v8
├── README.md
├── LICENSE.txt # detailed term of the software license
└── scripts
    ├── custom_YOLO.py # script which overwrites the default YOLO class
    ├── render_results.py  # helper function to annotate predictions
    ├── requirements.txt  # pip requirements
    └── test_script.py  # test script

Demonstration

Assuming you are in the scripts folder, you can run python3 test_script.py. This script saves the annotated image in ../images/annotated_1688033955437.jpg.

Make sure that the Python packages found in requirements.txt are installed. In case they are not, simply run pip3 install -r requirements.txt.

A live demonstration is freely accesible here.

Training metrics

The model has been trained using the custom YOLO class found in ./scripts/custom_YOLO.py. The custom YOLO class can be exactly used as the original YOLO class. The hyperparameters used during the training can be found in ./scripts/args.yaml.

The training maximize the [email protected], which is the mean Average Precision calculated at a 0.5 Intersection over Union (IoU) threshold, measuring how well the model detects objects with at least 50% overlap between predicted and ground truth bounding boxes.

Test metrics

<img src="images/F1_curve.png" width="300"> <img src="images/P_curve.png" width="300"> <img src="images/PR_curve.png" width="300"> <img src="images/R_curve.png" width="300">

License

Croppie cherry detection model © 2024 by Alliance Bioversity & CIAT, Producers Direct and M-Omulimisa is licensed under GNU-GPLv3

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/.

The detailed terms of the license are available in the LICENSE file in the repository.

Funding

Funded by: Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) Fair Forward Initiative - AI for All