Microstructure Image GAN

This project trains a Generative Adversarial Network (GAN) to synthesize realistic microstructure images based on input RGB image data. It supports multi-class training and saves both intermediate and final results for further analysis.

---

Installation Guide

Follow the steps below to set up the environment and run the project.

✅ Step 1: Install Miniconda

🔹 Windows

1. Download the Miniconda installer: Miniconda for Windows
2. Run the .exe installer and follow the prompts.
3. Open the Anaconda Prompt after installation.

🔹 macOS

# Download and install Miniconda
curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
bash Miniconda3-latest-MacOSX-x86_64.sh

Then restart your terminal.

🔹 Linux

# Download and install Miniconda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Then restart your shell or run: source ~/.bashrc

---

✅ Step 2: Create the Conda Environment

Clone the repository (or download the files) and make sure you're in the root directory.

git clone https://github.com/cmaloney111/deep-microstructure-reconstruction.git
conda env create -f environment.yml
conda activate MicroRec

---

✅ Step 3: Navigate to the Training Script

cd src/new-final-gan

---

✅ Step 4: Prepare Your Data

Your RGB image folder should be structured like this:

/path/to/RGB-image-folder/
├── class_0/
│   ├── image_0.jpg
│   ├── image_1.jpg
│   └── ...
├── class_1/
│   ├── image_0.jpg
│   ├── image_1.jpg
│   └── ...
└── ...

Each subfolder represents a class label (e.g., microstructure type), and contains images in .jpg/.jpeg, .png, or .tif format.

---

✅ Step 5: Run the Training Script

python train.py --path /path/to/RGB-image-folder

Arguments:

• --path: Path to the root folder containing your image classes.

Example:

python train.py --path ./data/microstructures

---

Outputs

After training, you will find:

• Generated images (images folder)
• Model checkpoints (models folder)