Fisheries Monitoring with Advanced Deep Learning

This project implements state-of-the-art computer vision models to detect and classify fish species from fishing boat camera footage.

🧠 Architectures

The project features two advanced architectures located in src/models/:

1. CNN (ResNet50) (src/models/cnn.py):

   • Uses a Pre-trained ResNet50 backbone.
   • Custom heads for Classification (8 classes) and Bounding Box Regression.
   • Robust and distinct feature extraction.

2. Transformer (ViT) (src/models/transformer.py):

   • Uses a Vision Transformer (vit_base_patch16_224) backbone.
   • Leverages self-attention mechanisms for global context.
   • Fine-tuned for simultaneous classification and localization.

📂 Project Structure

├── data/               # Dataset (Train images and JSON annotations)
├── app.py              # Gradio Web Application
├── src/
│   ├── train.py        # Main Training Script
│   ├── models/         # Neural Architectures
│   │   ├── cnn.py      # ResNet50 Implementation
│   │   └── transformer.py # Vision Transformer Implementation
│   └── ...
└── requirements.txt    # Dependencies

🚀 Getting Started

1. Installation

conda create -n fisheries python=3.10
conda activate fisheries
pip install -r requirements.txt

2. Training

Train your preferred model architecture.

Train CNN (ResNet):

python src/train.py --model_type cnn --epochs 10 --learning_rate 1e-4

Train Transformer (ViT):

python src/train.py --model_type vit --epochs 10 --learning_rate 5e-5

3. Web Application (Demo)

Launch the interactive web interface to test the model.

python app.py --model_type cnn --model_path src/models/best_model_cnn.pth

Note: Make sure to select the architecture that matches your trained model.

🌐 Publishing

This project uses Gradio for the web interface and can be automatically deployed to Hugging Face Spaces using the included GitHub Action.

Automated Deployment (GitHub Actions)

1. Create a New Space on Hugging Face (e.g., fisheries_monitoring).
2. In your GitHub Repository Settings -> Secrets and variables -> Actions:
   • Create a New Repository Secret named HF_TOKEN with your Hugging Face Access Token (Write permissions).
3. Edit .github/workflows/sync_to_hub.yml to match your Hugging Face username and space name if they differ from the default.
4. Push to main, and the action will automatically sync your code to the Space!

Manual Deployment

1. Create a Space on Hugging Face.
2. Upload app.py, src/, and requirements.txt.