FISHAI

FISHAI is a deep learning framework for fish image classification. It includes dataset enhancement, model training, and performance benchmarking against baseline architectures.

Project Structure

FISHAI/
│
├── Dataset/
│   ├── RawDataset/             # Unprocessed dataset
│   └── EnhancedDataset/        # Processed dataset ready for training
│
├── model_weight/
│   ├── baselines/              # Saved baseline model weights
│   ├── fish_cnn.pth            # Trained FISHAI CNN model
│   └── training_curves.png     # Performance curves
│
├── enhanced_processing.py      # Enhances raw dataset for training
├── cnn.py                      # CNN model definition
├── training.py                 # Training pipeline for the FISHAI model
├── baseline.py                 # Baseline model training script
├── __init__.py

1. Dataset Enhancement

Convert raw data into a structured and preprocessed format using:

python enhanced_processing.py --input_dir <path_to_RawDataset> --output_dir <path_to_EnhancedDataset>

2. CNN Model Setup

Define the FISHAI CNN architecture:

python cnn.py

3. Train FISHAI Model

Train the enhanced CNN model on the enhanced dataset:

python training.py \
  --num_classes <int> \
  --epochs <int> \
  --lr <float> \
  --weight_decay <float> \
  --batch_size <int> \
  --step_size <int> \
  --gamma <float> \
  --data_dir <path_to_EnhancedDataset> \
  --img_size <int> \
  --rotation <float> \
  --scale_min <float> \
  --scale_max <float> \
  --mean <mean_values> \
  --std <std_values> \
  --save_dir <output_model_dir> \
  --model_name <model_identifier> \
  --device <cuda_or_cpu>

4. Train Baseline Models

Run baseline comparisons using standard architectures:

python baseline.py \
  --num_classes <int> \
  --epochs <int> \
  --batch_size <int> \
  --data_dir <path_to_EnhancedDataset> \
  --img_size <int> \
  --mean <mean_values> \
  --std <std_values> \
  --save_dir <output_model_dir> \
  --device <cuda_or_cpu>

Output

• Trained model weights saved under model_weight/
• Training curves and performance metrics stored as training_curves.png

Please mention this repository in your work if you find it useful.