Generative AI Models (2AMU20) Assignment 3

This repository contains the third assignment for the Generative AI Models course (2AMU20) at Eindhoven University of Technology. The assignment focuses on converting low-resolution grayscale images into high-resolution colored images by combining two advanced generative models: conditional Generative Adversarial Networks (cGAN) for colorization and Super-Resolution GAN (SRGAN) for enhancing image resolution.

Key Components

1. Image Colorization: Utilizing cGAN to colorize grayscale images.
2. Super-Resolution: Utilizing SRGAN to enhance the resolution of images.
3. Integration: Combining both models to transform 64x64 grayscale images into 256x256 high-resolution colored images.

Training Data

• Colorization Model: Trained on a subset of the COCO dataset comprising 10,000 images.
• Super-Resolution Model: Trained on the PASCAL dataset using 10,000 images.

How to Run

1. Clone the repository:

   git clone https://github.com/luca-mainardi/Image-Colorization-and-SuperResolution.git

2. Navigate to the project directory:

   cd Image-Colorization-and-SuperResolution/

3. Install the required dependencies:

   pip install -r requirements.txt

4. Follow the instructions in the provided Jupyter notebooks to set up the environment and run the code.
   You can download the weights of the models at this link: https://drive.google.com/drive/folders/1dOVFTqZ9uaPBzkVa-gVxZgGIEZy9m125?usp=sharing.
   It is recommended to run the project on Google Colab or kaggle.

Contributors

• Luca Mainardi
• Francesco Brescia
• Matthew Nana