Abstract:Diffusion models have gained significant popularity in the field of image-to-image translation. Previous efforts applying diffusion models to image super-resolution (SR) have demonstrated that iteratively refining pure Gaussian noise using a U-Net architecture trained on denoising at various noise levels can yield satisfactory high-resolution images from low-resolution inputs. However, this iterative refinement process comes with the drawback of low inference speed, which strongly limits its applications. To speed up inference and further enhance the performance, our research revisits diffusion models in image super-resolution and proposes a straightforward yet significant diffusion model-based super-resolution method called ACDMSR (accelerated conditional diffusion model for image super-resolution). Specifically, our method adapts the standard diffusion model to perform super-resolution through a deterministic iterative denoising process. Our study also highlights the effectiveness of using a pre-trained SR model to provide the conditional image of the given low-resolution (LR) image to achieve superior high-resolution results. We demonstrate that our method surpasses previous attempts in qualitative and quantitative results through extensive experiments conducted on benchmark datasets such as Set5, Set14, Urban100, BSD100, and Manga109. Moreover, our approach generates more visually realistic counterparts for low-resolution images, emphasizing its effectiveness in practical scenarios.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper primarily addresses two key issues in Single Image Super-Resolution (SISR): 1. **Limitations of existing methods**: - Traditional diffusion model-based methods can generate high-quality images in super-resolution tasks, but their iterative refinement process results in slow inference speed, limiting practical applications. - Generative Adversarial Network (GAN) methods can produce realistic images, but they are prone to mode collapse during training, and the adversarial loss often introduces artifacts not present in the original image, leading to significant distortion. 2. **Proposed new method**: - The paper proposes an Accelerated Conditional Diffusion Model for Image Super-Resolution (ACDMSR), aiming to improve the quality and inference speed of super-resolution images through a deterministic iterative denoising process. - This method utilizes a pre-trained super-resolution model to generate conditional images and combines it with a diffusion model for iterative denoising, significantly reducing the inference steps from 1000 to 40 while ensuring image quality. ### Main Contributions - **Combining diffusion model with pre-trained super-resolution model**: For the first time, the diffusion model is combined with existing pre-trained super-resolution models as a post-processing framework for super-resolution tasks. - **Deterministic sampling method**: A deterministic sampling method is adopted, effectively reducing the sampling steps during inference, achieving better perceptual quality and lower distortion. - **Outstanding experimental results**: Achieved superior results compared to existing state-of-the-art methods on multiple standard benchmark datasets (such as Set5, Set14, Urban100, BSD100, and Manga109), and generated more realistic super-resolution images.

ACDMSR: Accelerated Conditional Diffusion Models for Single Image Super-Resolution

ACDMSR: Accelerated Conditional Diffusion Models for Single Image Super-Resolution

CDPMSR: Conditional Diffusion Probabilistic Models for Single Image Super-Resolution

AdaDiffSR: Adaptive Region-aware Dynamic Acceleration Diffusion Model for Real-World Image Super-Resolution

A Conditional Diffusion Model With Fast Sampling Strategy for Remote Sensing Image Super-Resolution

Implicit Diffusion Models for Continuous Super-Resolution

Denoising Diffusion Probabilistic Model with Adversarial Learning for Remote Sensing Super-Resolution

SRDiff: Single image super-resolution with diffusion probabilistic models

Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution

Single image super-resolution with denoising diffusion GANS

Enhancing Remote Sensing Image Super-Resolution with Efficient Hybrid Conditional Diffusion Model

HSR-Diff:Hyperspectral Image Super-Resolution via Conditional Diffusion Models

Latent Diffusion, Implicit Amplification: Efficient Continuous-Scale Super-Resolution for Remote Sensing Images

HSR-Diff: Hyperspectral Image Super-Resolution Via Conditional Diffusion Models

ResDiff: Combining CNN and Diffusion Model for Image Super-Resolution

Diffusion Models for Image Super-Resolution: State-of-the-art and Future Directions

ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting

AddSR: Accelerating Diffusion-based Blind Super-Resolution with Adversarial Diffusion Distillation

Improving the Stability and Efficiency of Diffusion Models for Content Consistent Super-Resolution

TASR: Timestep-Aware Diffusion Model for Image Super-Resolution