IMAGE PROCESSING METHOD AND DEVICE, NEURAL NETWORK TRAINING METHOD, IMAGE PROCESSING METHOD OF COMBINED NEURAL NETWORK MODEL, CONSTRUCTION METHOD OF COMBINED NEURAL NETWORK MODEL, NEURAL NETWORK PROCESSOR AND STORAGE MEDIUM

摘要

An image processing method and device, a neural network training method, an image processing method of a combined neural network model, a construction method of a combined neural network model, a neural network processor, and a storage medium. The image processing method comprises: obtaining initial feature images of N levels ordered from high to low resolution on the basis of an input image, N being a positive integer, and N>2; performing cyclic scaling processing on the initial feature image of the first level on the basis of the initial feature images of 2-N levels to obtain intermediate feature images; and synthesizing the intermediate feature images to obtain an output image. The cyclic scaling processing comprises: nested scaling processing of N-1 levels; the scaling processing of each level comprises down-sampling processing, linkage processing, up-sampling processing, and residual link addition processing that are executed in sequence; linkage is performed on the basis of the output of the down-sampling processing of the current level and the initial feature image of the next level in the linkage processing of the current level to obtain the output of the joint processing of the current level; and the scaling processing of the next level is nested between the down-sampling processing and the joint processing of the current level.