GANUnet-unblur: GAN-based Image Restoration Using U-Net

GANUnet-unblur is a project that uses Generative Adversarial Networks (GAN) for image restoration. Specifically, the generator is based on a U-Net architecture, which is widely used for tasks like image segmentation and restoration. The discriminator is based on a Convolutional Neural Network (CNN), helping distinguish between real sharp images and the blurred images restored by the generator.

Requirements

• Python 3.7+
• PyTorch
• Torchvision
• Matplotlib
• PIL
• Other Python dependencies (listed in requirements.txt)

Installation

Clone the repository and install the required dependencies:

git clone https://github.com/AjitKumar-git/GANUnet-unblur.git
cd GANUnet-unblur
pip install -r requirements.txt

Dataset Loader

The dataset loader handles different types of datasets, including loading paired images from a single folder, two separate folders, or real-world blurred images for training. The loader allows you to easily load and visualize the dataset.

Available Dataset Types

1. One Folder Dataset
   This type assumes that all images are located in a single folder, where each image has a corresponding sharp or ground truth version. It loads both blurred and sharp images from a single folder.

2. Two Folder Dataset
   This type loads images from two separate folders. One folder contains the blurred images, while the other contains the corresponding sharp images. The images are paired by their filenames.

3. Real Blur Dataset
   This type is used for real-world blurred datasets, where the dataset contains pairs of blurred and sharp images in different folders (e.g., Scenes where blur and gt folders are present).

Using the Dataset Loader

You can use the dataset loader to load datasets of different types, and optionally visualize a few sample images. Below is an example of how to load and visualize the dataset.

Example Usage:

For RealBlur Dataset:

python main.py --mode train --dataset_type real_blur --root_dir "your_realblur_dataset_location(root directory only)" --image_size "256,256" --num_epochs 100

For One Folder Dataset:

python main.py --mode train --dataset_type one_folder --root_dir "your_dataset_location(root directory only)" --image_size "256,256" --num_epochs 100

For Two Folder Dataset:

python main.py --mode process --dataset_type two_folder --blurred_folder blur --sharp_folder sharp --image_size "256,256" --epoch 50

---