Single-Image Super-Resolution with Deep Learning

Enhance the resolution and visual quality of low-resolution images using a deep learning-based Single-Image Super-Resolution system. This project leverages Convolutional Neural Networks (CNNs) to intelligently upscale images, providing a high-resolution output that preserves fine details and improves overall image quality.

Table of Contents

• Getting Started
  • Prerequisites
  • Installation
• File Structure
• Output

Getting Started

Data sets and softwares to be installed are provided below to set up and deploy the deep learning-based Super-Resolution system on your local machine.

Prerequisites

Download datasets from below links:
DIV2K -- https://www.kaggle.com/datasets/joe1995/div2k-dataset.

Urban100 --https://www.kaggle.com/datasets/jesucristo/super-resolution-benchmarks.

We have Utilized DIV2k for training the models and Urban100 for testing.

Installation

You will need the following to run files:

Python

tensorflow

keras

numpy

tqdm

matplotlib, skimage, scipy

File Structure

ISR_VDSR.py : Contains preprocessing of the data and models with basic ISR with CNN and VDSR.

ISR_EDSR.py : Contains EDSR model.

Output

Below are few results