The repository Sajid-Bit/GAN-Doc-Quality focuses on improving the quality of documents using Generative Adversarial Networks (GANs). It includes the implementation, datasets, and results of the project aimed at enhancing document image quality.
The data folder contains two subdirectories:
img_A: This folder contains images with low quality.img_B: This folder contains images with high quality.
The GAN model used in this project consists of two main components:
- Generator: This neural network generates high-quality images from noisy or low-quality images. It tries to fool the discriminator by producing images that look similar to high-quality images.
- Discriminator: This neural network distinguishes between high-quality images and the images generated by the generator. It is trained to correctly identify real high-quality images and reject the fake images produced by the generator.
The training process involves the generator and discriminator competing against each other, with the generator improving its ability to create realistic images, and the discriminator getting better at detecting fake images.
The project includes a notebook that performs the following tasks:
- Adds noise to images.
- Trains the GAN model to improve the quality of noisy images.
The test folder contains the code for testing the trained GAN model. It includes scripts for:
- Loading the trained model.
- Running the model on test images.
- Evaluating the quality of the generated images compared to the high-quality images.