Training method and device of image super-division model and computer readable storage medium

Abstract

The invention provides a training method and device for an image super-division model. The training method comprises the steps of forming an initial model which comprises a super-division module composed of a plurality of information distillation sub-modules which are sequentially connected end to end, wherein the information distillation sub-module comprises a plurality of residual units which are sequentially connected end to end; inputting the low-resolution image into the initial model, and extracting shallow layer features of the low-resolution image; extracting the deep features from the shallow features through a super-division module; summing and fusing the shallow-layer features and the deep-layer features, and then performing up-sampling processing to obtain a super-resolution image output by the initial model; and training the initial model according to a loss function determined by the super-resolution image and the high-resolution image to obtain an image super-resolution model. According to the method, refined features can be extracted by using an information multi-distillation structure based on a residual unit, and superposing shallow features with the refined features, so that the problems of degradation and gradient generated when a model network is in a deep layer are solved, and the parameter quantity and the calculation quantity are greatly reduced.

Description

technical field [0001] The present application belongs to the field of computer technology, and in particular relates to a training method and device for an image super-resolution model, an image super-resolution method and device, and a computer-readable storage medium. Background technique [0002] Functions such as special effect cameras and shooting props have become standard configurations for video applications. However, due to the limitation of equipment computing power, the on-end model only allows inference at a small image resolution, resulting in blurred images. Therefore, super resolution is required. Algorithms to improve the overall image effect. [0003] In related technologies, a trained convolutional neural network model can be used to perform super-resolution processing on an original image to obtain a super-resolution image with higher definition. [0004] However, in the current solution, the convolutional neural network model has a large amount of calcu...

AI Technical Summary

Problems solved by technology

[0005] In view of this, this application provides a training method, device, and computer-readable storage medium for an image super-resolution model, which solves the calculation, parameter, and memory read / write volumes of the convolutional neural network model in the current solution to a certain extent. They are all very large, and it is difficult to deploy for devices with weak computing power, which makes it difficult for the response time to meet the real-time requirements

Images