metadata
library_name: pytorch
license: mit
pipeline_tag: image-to-image
tags:
- pytorch
- computer-vision
- image-super-resolution
- diffusion
LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction
This repository contains the official model weights for LPNSR, a prior-enhanced efficient diffusion framework for image super-resolution (SR).
- Paper: LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction
- GitHub Repository: Faze-Hsw/LPNSR
Introduction
LPNSR addresses the efficiency-quality trade-off in diffusion-based SR. While state-of-the-art frameworks like ResShift achieve efficient 4-step inference, they can suffer from performance degradation due to unconstrained random noise. LPNSR addresses this by:
- Replacing random Gaussian noise with an LR-guided multi-input-aware noise predictor, embedding structural priors into the reverse process.
- Mitigating initialization bias using a high-quality pre-upsampling network to optimize the diffusion starting point.
- Maintaining a compact 4-step sampling trajectory for high-quality, real-world super-resolution.
Features
- Efficient Sampling: Only 4 sampling steps required for high-quality super-resolution.
- Noise Predictor: Learns to predict optimal noise maps for partial diffusion initialization.
- Real-world SR: Designed to handle complex real-world degradations.
- SwinIR Integration: Optional SwinIR refinement for enhanced details.
Quick Start
To use these weights, clone the official repository and follow the environment setup instructions.
Inference
Once the environment is set up and weights are placed in the pretrained/ folder, run:
python LPNSR/inference.py -i [image folder/image path] -o [output folder]
Online Demo
You can also launch a local Gradio web interface:
python LPNSR/app.py
Citation
If you find this work useful, please cite:
@article{lpnsr2026,
title={LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction},
author={Huang, Shuwei and Liu, Shizhuo and Wei, Zijun},
journal={arXiv preprint arXiv:2603.21045},
year={2026},
eprint={2603.21045},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
Acknowledgement
This project is based on ResShift, BasicSR, SwinIR, and Real-ESRGAN.