savingface: Humphead Wrasse Individual Identification

This repository contains the open-source code for training the artificial intelligence models described in the paper:

Leveraging artificial intelligence for photo identification to aid CITES enforcement in combating illegal trade of the endangered humphead wrasse (Cheilinus undulatus)
C. Y. Hau, W. K. Ngan, Y. Sadovy de Mitcheson
Frontiers in Ecology and Evolution, 2025
DOI: 10.3389/fevo.2025.1526661
Link: https://www.frontiersin.org/journals/ecology-and-evolution/articles/10.3389/fevo.2025.1526661/full

Models

This repository provides the code for training two key deep learning models:

1. Facial Pattern Extraction: A YOLOv8-based model trained to detect and extract the left and right facial patterns from images of humphead wrasse. The model achieved a 99% success rate in the study.
2. Individual Identification: A ResNet-50-based convolutional neural network (CNN) retrained using a triplet loss function. This model compares extracted facial patterns to identify individual fish.

Getting Started

(Placeholder: Add instructions on setting up the environment, installing dependencies (e.g., requirements.txt), and preparing the necessary datasets.)

# Example dependency installation
pip install -r requirements.txt

Training

(Placeholder: Add detailed instructions on how to run the training scripts for both the extraction and identification models. Include necessary command-line arguments or configuration file details.)

# Example training command for YOLOv8 extraction model
# The training process uses an IPython Notebook (.ipynb), which can be run directly in environments like Google Colab.
# Refer to the notebook for specific instructions.

# Example training/evaluation command for ResNet-50 identification model
python train/evaluation/triplet_loss/train.py

Usage

(Placeholder: Add instructions on how to use the trained models for inference, either via provided scripts or by integrating them into other applications like the 'Saving Face' app.)

Citation

If you use this code or the associated models in your research, please cite the original paper:

@article{hau2025savingface,
  title={Leveraging artificial intelligence for photo identification to aid CITES enforcement in combating illegal trade of the endangered humphead wrasse ({_Cheilinus undulatus_})},
  author={Hau, C. Y. and Ngan, W. K. and Sadovy de Mitcheson, Y.},
  journal={Frontiers in Ecology and Evolution},
  volume={13},
  year={2025},
  pages={1526661},
  doi={10.3389/fevo.2025.1526661}
}

Contact

For questions about the research and methodology, please refer to the contact information in the original publication.