Yak image generation method based on style conversion

Abstract

The invention discloses a yak image generation method based on style conversion, and relates to the technical field of computer image generation. The yak image generation method based on style conversion comprises the following steps that S1, a yak image and an image with extracted yak semantic information are used as minimum unit components of a yak data set, and the image with extracted yak semantic information and feature vectors are transmitted into a network together; the network generation complexity is reduced, and the result accuracy is improved. And meanwhile, a loss function of a traditional generative adversarial network is improved, so that the network tends to protect the generation effect of the yak image in training.

Description

technical field [0001] The invention relates to the technical field of computer image generation, in particular to a yak image generation method based on style conversion. Background technique [0002] At present, scientific research mainly focuses on designing deep neural networks suitable for yak images. It can be roughly divided into two categories: one is to establish the mapping relationship between the feature vector and the output image or other types of information, which is mainly used to complete tasks such as image generation and data enhancement; the other is to establish the mapping between the input image and the output image. relationship, mainly used to complete tasks such as style transfer, super-resolution reconstruction, and image inpainting; [0003] There are not many existing research cases for yak image generation. The traditional generative adversarial network training initialization feature vector is too complicated, it is difficult to accurately fi...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the deficiencies of the prior art, the present invention provides a yak image generation method based on style conversion, which solves the problem that there are not many research cases on yak image generation, and the traditional GAN ​​training initialization feature vector is too complicated and difficult to solve. It is difficult to accurately fit the outline of the yak, and the details are too rough. With the increase of the number of labels, the network has problems such as model collapse, and for the incomplete data set, the decoding of yak images is too complicated, and the feature entanglement is too complicated to learn. Problems with generated mapping functions

Images