MDFN: Efficient Image Super-Resolution through Multi-Domain Feature Fusion

📄 Paper Accepted: ICVGIP 2025

This repository is the official implementation of our paper, "MDFN: Efficient Image Super-Resolution through Multi-Domain Feature Fusion".

🚧 Code Release Status

Thank you for your interest in our work! We are currently cleaning and organizing the code to make it easy to use and reproduce.

The code will be released soon. Please "Watch" or "Star" this repository to be notified of the official release.

💡 Abstract

We introduce the Multi-Domain Feature Network (MDFN), a novel and resource-efficient architecture for single image super-resolution. MDFN synergistically fuses features from three complementary domains:

• Spatial Domain: Using convolutions for local textures.
• Multi-Scale Domain: Using a Laplacian pyramid for hierarchical features.
• Frequency Domain: Using Fourier transforms for global context.

⚙️ Coming Soon

• Full source code for training and testing.
• Pre-trained models for standard benchmarks (e.g., Set5, Set14, BSD100).
• Example inference scripts.
• Detailed instructions for reproducing our paper's results.