THE METHOD AND SYSTEM FOR DESIGNING DEPTH CONTROLLABLE NEURAL NETWORKS FOR SUPER RESOLUTION RECONSTRUCTION

摘要

The present invention relates to a method and system for designing a neural network capable of adjusting depth for ultra-high resolution restoration, comprising: a neural network design method having an input layer, an output layer, and a plurality of hidden layers, wherein the neural network design system cascades the layers in a neural network. ) Training; Copying, by the neural network design system, a shallowest layer among the plurality of hidden layers from a pre-learned layer; Initializing, by the neural network design system, the remaining layers not copied from the pre-learned layer among the plurality of hidden layers; The neural network design system may include adding an output branch in the middle of the hidden layer.