Medfusion Fake Image Generator

Medfusion Fake Image Generator is a project focused on generating synthetic medical images using Variational Autoencoders (VAE) and Diffusion Models. The primary goal is to improve medical image analysis through high-quality image generation, particularly for datasets with limited data. The project includes pre-trained models, custom data processing scripts, and support for integrating new datasets.

---

Features

• Customizable Medical Image Generation: Generate high-quality synthetic images using advanced models.
• Pretrained VAE and Diffusion Models: Models optimized for datasets like BUSI and CheXpert.
• Supports 2D and 3D Data Augmentation: Enhance your dataset with various augmentation techniques.
• Flexible Dataset Integration: Easily integrate your own datasets with customizable configurations.

Installation

1. Clone the Repository:

   git clone https://github.com/Steven-ZN/Medfusion_Fake_Image.git
   cd Medfusion_Fake_Image

2. Install Dependencies: Ensure you have Python 3.8+ and pip installed. Run:

   pip install -r requirements.txt

3. Set Up Environment:

   • Configure GPU and CUDA environment for faster training and inference.
   • Ensure you have at least one NVIDIA GPU with CUDA 11+ installed.

Usage

1. Training the VAE

Train the Variational Autoencoder (VAE) on your dataset:

python scripts/train_latent_embedder_2d.py --config configs/vae_config.yaml

2. Generating Images with Diffusion Model

Once the VAE is trained, use it to train the diffusion model for image generation:

python scripts/train_diffusion.py --config configs/diffusion_config.yaml

3. Evaluate the Model

Evaluate generated images and embeddings:

python scripts/evaluate_images.py --checkpoint path/to/vae_checkpoint.pth

---

Dataset Configuration

Preprocessing the BUSI Dataset

1. Place your dataset under data/ with the following structure:

   data/
   ├── train/
   │   ├── benign/
   │   ├── malignant/
   │   ├── normal/
   ├── validation/
   │   ├── benign/
   │   ├── malignant/
   │   ├── normal/
   ├── test/
   │   ├── benign/
   │   ├── malignant/
   │   ├── normal/
   

2. Run preprocessing:

   python scripts/helpers/sample_dataset.py --input data/ --output processed_data/

Custom Dataset Integration

1. Add your dataset to the data/ folder.
2. Modify the dataset loader in medical_diffusion/data/datasets to include your data parsing logic.
3. Update configurations in configs/.

---

Customization

Adjusting Input Size

To change the image resolution:

1. Update the img_size parameter in the VAE and diffusion model configuration files (configs/vae_config.yaml and configs/diffusion_config.yaml).
2. Example:

   img_size: 512

Using a Custom Optimizer

The project supports custom optimizers. Modify the optimizer settings in the configuration file:

optimizer:
  type: CustomOptimizer
  params:
    lr: 0.001
    weight_decay: 0.0001

---

Contributing

We welcome contributions! Feel free to:

1. Open issues for bugs or feature requests.
2. Submit pull requests with enhancements or fixes.

---

Acknowledgments

This project uses concepts and resources from:

• BUSI and CheXpert datasets.
• Variational Autoencoders (VAEs) and Diffusion Models.

---

License

This project is licensed under the MIT License. See the LICENSE file for more details.

---

Contact

For questions or collaborations, please reach out via GitHub Issues.

This README provides a complete overview, setup instructions, and customization details for your project. Let me know if you need further refinements!
![Medfusion](https://github.com/user-attachments/assets/2b50a57c-cfa2-40cd-9b9e-abd84649ad99)