itznoor998/SwinIR-4x-SuperResolution

SwinIR 4x Super-Resolution

This repository contains a fine-tuned SwinIR model for 4x image super-resolution.

Model

• Architecture: SwinIR

• Task: Image Super-Resolution

• Scale factor: 4x

• Input: RGB low-resolution image

• Output: RGB high-resolution image

Architecture

• Embed dimension: 180
• Depths: [6, 6, 6, 6, 6, 6]
• Number of heads: [6, 6, 6, 6, 6, 6]
• Window size: 8
• Upsampler: PixelShuffle
• Residual connection: 1conv

Usage

Install dependencies:

pip install torch torchvision pillow

Then:

python inference.py

Training

The model was initialized from the official SwinIR 4x classical super-resolution model and fine-tuned on the project dataset.