Nested loop consistency-based generative adversarial network image style transfer method

Abstract

The invention belongs to the field of image processing, computer vision and deep learning, and specifically discloses a nested loop consistency-based generative adversarial network image style transfer method. According to the method, mapping, from content images such as photos and the like to claborate-style painting, of style images is realized on the basis of a convolutional neural network, a generator, a discriminator in residual connection with the generator, and nested loop consistency and generative adversarial network-based loss function training. The method is capable of effectively solving image style transfer tasks which cover geometric changes; and moreover, the method does not need to pair data sets in a one-to-one correspondence manner, and is capable of learning mapping fromlearning content pictures to style pictures and the mapping from style pictures to content pictures at the same time.

Description

technical field [0001] The invention belongs to the fields of image processing, computer vision, and deep learning, and specifically relates to a method for generating confrontational network image style transfer based on multiple cycle consistency. Background technique [0002] In recent years, along with the advances in deep learning, image style transfer techniques have also undergone important developments. In 2016, Leon A. Gatys published the paper "Image Style Transfer Using Convolutional NeuralNetworks", using deep learning algorithms to transfer image styles. The principle behind it is to use convolutional neural networks to perform content and style features at different scales. separation, thus making image style transfer easy and feasible. In 2017, Jun-Yan Zhu published the paper "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks", which uses cycle consistency and generative confrontational networks for image-to-image mapping learnin...

AI Technical Summary

Problems solved by technology

In style transfer tasks involving color and texture changes, these methods usually achieve good results, but in the face of tasks involving geometric changes, these methods often perform poorly, such as converting photos into Chinese paintings, landscape paintings, and meticulous paintings. other style pictures

Images