This repository contains an implementation of a Deep Convolutional Generative Adversarial Network (DCGAN) trained on the OASIS dataset. The model is designed to generate synthetic brain scan images based on the dataset provided.
Generative Adversarial Networks (GANs) consist of two neural networks, a generator and a discriminator, that are trained simultaneously through adversarial learning:
- The generator (G) takes in a random noise vector and generates fake images that resemble the real dataset.
- The discriminator (D) evaluates images and tries to distinguish between real and fake images.
- The generator is optimized to fool the discriminator, while the discriminator is optimized to correctly classify images.
DCGANs extend this concept using convolutional layers to improve the quality of generated images.
- Uses transposed convolutional layers to upsample random noise into an image.
- Includes batch normalization and ReLU activation for stable training.
- Outputs images using a Tanh activation function.
- Uses convolutional layers with LeakyReLU activation to classify images as real or fake.
- Includes batch normalization for improved gradient flow.
- Outputs a probability score using a Sigmoid activation function.
- Python 3.12+
- PyTorch
- Torchvision
- Matplotlib
- NumPy
Run the training script:
python torch_dcgan.py
After training, the model will generate synthetic images. The results are saved as:
brains.jpg: Comparison of real vs. generated images.loss.jpg: Training loss curves for generator and discriminator.
- Radford, A., Metz, L., & Chintala, S. (2015). Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks.
- OASIS Dataset: https://www.oasis-brains.org/
- PyTorch DCGAN Tutorial: https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html
MIT License. Feel free to modify and use the code!