Semantic Image Synthesis for Dental Panoramic X-ray Images

This repository contains the implementation of OASIS (Only Adversarial Segmentation-to-Image Synthesis) and GauGAN (SPADE) for generating realistic dental panoramic X-ray images from semantic segmentation masks.

🚀 Features

• Semantic-to-Image Synthesis: Generate high-fidelity dental X-rays from multi-class labels.
• Dual Model Support: Implements both OASIS and GauGAN architectures.
• Medical Imaging Focus: Specifically tuned for the TUFTS Dental Database and dental panoramic image characteristics.
• Evaluation Suite: Includes metrics such as SSIM, FID, and clinical validation tools.
• Interactive App: A Streamlit-based application for real-time inference and testing.

🛠️ Installation

1. Clone the repository:

   git clone https://github.com/SerdarHelli/SemanticImageSynthesisDentalPanoramic.git
   cd SemanticImageSynthesisDentalPanoramic

2. Install dependencies:

   pip install -r requirements.txt

📈 Training

To train the model, use the train.py script. You need to provide the paths to the training and validation data.

python train.py --train_data_path /path/to/train --val_data_path /path/to/val --model oasis --epochs 50 --batch_size 1

Key Arguments:

• --model: Choose between oasis or gaugan.
• --img_size: Input image size (default 256).
• --include_abnormality: Boolean to include or exclude abnormality classes (default True).
• --logs_path: Directory to save checkpoints and TensorBoard logs.

🧪 Evaluation

Run the evaluation script to calculate metrics on your test set:

python eval.py --test_data_path /path/to/test --model_path /path/to/checkpoint

🌐 Web Application

Launch the interactive Streamlit app for testing:

streamlit run app/fisher_test_app.py

🧠 Architectures

• OASIS: A segmentation-to-image synthesis model that uses a discriminator trained to perform semantic segmentation, eliminating the need for a separate perceptual loss in some configurations.
• GauGAN (SPADE): Utilizes Spatially-Adaptive Normalization to better preserve semantic information throughout the generator.

🏷️ GitHub Keywords

semantic-image-synthesis, dental-panoramic, gan, oasis-gan, gaugan, spade, tensorflow, keras, medical-imaging, deep-learning, image-generation, tufts-dental-database, x-ray-synthesis, computer-vision

📄 License

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