MageAI - Image Upscaling with ESRGAN

MageAI is an advanced image upscaling web application powered by the Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN). This application allows users to upload low-resolution images and receive high-resolution upscaled versions, leveraging state-of-the-art AI techniques.

Table of Contents

• Overview
• Features
• Technologies
• Installation
• Usage
• Model Architecture
• References

---

Overview

MageAI is designed for image upscaling and enhancement, specifically aimed at restoring details from low-resolution images. It uses ESRGAN (Enhanced Super-Resolution GAN) to improve image quality significantly by learning how to add high-frequency details lost in low-resolution images.

ESRGAN has revolutionized image super-resolution by introducing Residual-in-Residual Dense Blocks (RRDB) and the Relativistic GAN framework. Read more about ESRGAN here.

---

Features

• Upload: Upload an image (supports PNG, JPG, SVG, GIF).
• Upscale: Uses ESRGAN for image upscaling.
• Preview: View the upscaled image instantly.
• Download: Download the upscaled image.

---

Technologies

This project is built with:

• Frontend: React and Next.js
• Styling: Tailwind CSS
• Backend: FastAPI for handling image processing.
• Machine Learning Model: ESRGAN for image upscaling.
• Image Handling: Python's PIL library for image processing.
• Deployment: Render for hosting both frontend and backend services.

---

Installation

To run this project locally, follow the steps below:

Prerequisites

• Node.js (>= 16.x.x)
• Python (>= 3.9)
• FastAPI
• Install ESRGAN

Clone the repository

git clone https://github.com/PranavBhatP/Mage.ai.git
cd Mage.ai

Frontend Setup

1. Navigate to the client/ folder:

   cd client

2. Install dependencies:

   npm install

3. Start the development server:

   npm run dev

Backend Setup

1. Navigate to the server/ folder:

   cd server

2. Install dependencies (preferably in a virtual environment):

   pip install -r requirements.txt

3. Run the FastAPI backend:

   uvicorn main:app --host 0.0.0.0 --port 8000

---

Usage

1. Upload an Image: Upload your image (JPG, PNG, or GIF format).
2. Upscale: Click the "Let's Go!" button to process the image using the backend ESRGAN model.
3. Download the Image: Once processed, download your upscaled image.

---

Model Architecture

The backbone of this project is ESRGAN which improves upon the classic GAN by using:

• RRDB (Residual-in-Residual Dense Block): Enables better learning and stability for high-res image generation.
• Relativistic GAN: Helps the model better differentiate between real and generated high-resolution images.

Learn more about the architecture in the original ESRGAN paper.

---

References

• ESRGAN GitHub: https://github.com/xinntao/ESRGAN
• Original ESRGAN Paper: ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks
• FastAPI Documentation: https://fastapi.tiangolo.com/
• Next.js Documentation: https://nextjs.org/docs

---

License

This project is licensed under the MIT License - see the LICENSE file for details.