CycleGAN for Fluorescent Image Generation

This project implements a CycleGAN model for generating fluorescent microscopy images from synthetic masks. The model learns bidirectional mappings between synthetic segmentation masks and real fluorescent images.

Overview

CycleGAN (Cycle-Consistent Adversarial Networks) enables unpaired image-to-image translation. In this implementation:

• Domain A: Single-channel synthetic masks/segmentation images
• Domain B: Single-channel real fluorescent microscopy images

The model learns to:

1. Generate realistic fluorescent images from synthetic masks (A → B)
2. Generate masks from fluorescent images (B → A)
3. Maintain cycle consistency (A → B → A and B → A → B)

Features

• PyTorch implementation of CycleGAN
• Single-channel image support (grayscale masks ↔ grayscale fluorescent)
• Support for fluorescent microscopy image generation
• Automatic grayscale conversion for input images
• Customizable network architectures
• Training and inference scripts
• Data preprocessing utilities
• Tensorboard logging
• Model checkpointing

Requirements

• Python: 3.8 - 3.13 (tested with Python 3.13.7)
• CUDA: Optional but recommended for GPU acceleration
• Memory: At least 8GB RAM (16GB+ recommended for training)

Installation

# Clone the repository
git clone https://github.com/yourusername/cycleGAN.git
cd cycleGAN

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Note: This implementation has been tested with Python 3.13.7 and PyTorch 2.8.0. All dependencies install successfully.

Data Structure

data/
├── trainA/          # Synthetic masks for training
├── trainB/          # Fluorescent images for training
├── testA/           # Synthetic masks for testing
└── testB/           # Fluorescent images for testing

Usage

## Model Architecture

- **Generator**: ResNet-based architecture with skip connections
- **Discriminator**: PatchGAN discriminator
- **Loss Functions**: 
  - Adversarial loss
  - Cycle consistency loss
  - Identity loss (optional)

## Configuration

Modify `config.py` to adjust:
- Network architectures
- Training hyperparameters
- Data preprocessing settings
- Loss function weights

## Results

The model generates realistic fluorescent images that maintain structural consistency with input masks while adding realistic fluorescent characteristics.

## License

This project is licensed under the MIT License - see the LICENSE file for details.