Derivation of Eq. (6): According to Bayes's theorem, we have q ( x Please zoom in for a better view. The blurring kernel is randomly sampled from the isotropic Gaussian and anisotropic Gaussian kernels with a probability of [0.6, 0.4]. For isotropic Gaussian kernel, the kernel width is uniformly sampled from [0.2, 0.8]. We first added Gaussian and Poisson noise with a probability of [0.5, 0.5]. For Gaussian noise, the noise level is randomly chosen from [1,15].

Diffusion-based image super-resolution (SR) methods are mainly limited by the low inference speed due to the requirements of hundreds or even thousands of sampling steps. Existing acceleration sampling techniques inevitably sacrifice performance to some extent, leading to over-blurry SR results. To address this issue, we propose a novel and efficient diffusion model for SR that significantly reduces the number of diffusion steps, thereby eliminating the need for post-acceleration during inference and its associated performance deterioration.

Diffusion-based image super-resolution (SR) methods are mainly limited by the low inference speed due to the requirements of hundreds or even thousands of sampling steps. Existing acceleration sampling techniques inevitably sacrifice performance to some extent, leading to over-blurry SR results. To address this issue, we propose a novel and efficient diffusion model for SR that significantly reduces the number of diffusion steps, thereby eliminating the need for post-acceleration during inference and its associated performance deterioration. Additionally, an elaborate noise schedule is developed to flexibly control the shifting speed and the noise strength during the diffusion process. Extensive experiments demonstrate that the proposed method obtains superior or at least comparable performance to current state-of-the-art methods on both synthetic and real-world datasets, \textit{\textbf{even only with 20 sampling steps}}.