Acceleration Method for Super-Resolution Based on Diffusion Models by Intermediate Step Prediction

In this paper, we propose a new method to improve the generation speed of single-image super-resolution models based on the diffusion model. We address the generation speed problem of super-resolution models based on the diffusion model and propose an acceleration method by predicting intermediate steps. The proposed method is highly compatible with other sampling acceleration methods while maintaining high image quality and improving the efficiency and quality of the super-resolution task.