Semi-supervised multi-modal multi-class image translation method

Abstract

The invention discloses a semi-supervised multi-modal multi-category image translation method. The method comprises the following steps of S1, inputting two images from different domains and a small number of labels; S2, inputting the input images and labels into the encoders, dividing the encoders into a content encoder and a style encoder, and respectively decoupling the images from the style encoder and the content encoder by using the decoupling representation learning to obtain the style codes and the content codes; S3, inputting the style codes into an adversarial auto-encoder to complete the image multi-class training; inputting the content codes into a content adversarial learning network to complete the image multi-mode transformation training; and S4, realizing the image reconstruction and multi-mode conversion through splicing the style codes and the content codes. According to the method, the problem caused by the requirement for image translation diversity is solved, and the multi-modal and multi-class cross-domain images can be generated by jointly decoding the potential content codes and the style codes.

Description

technical field [0001] The present invention relates to the technical fields of computer vision, computer graphics and machine learning, in particular to a semi-supervised, multi-modal and multi-category image translation method. Background technique [0002] With the continuous development of deep learning technology and image generation technology, a large number of excellent works have emerged in the field of image translation. Semi-supervised multi-modal and multi-category image translation is an important and challenging research problem in the field of computer vision. Semi-supervised multi-modal Stateful multi-category image translation has obvious application value in industrial design and other fields, and can be applied to various aspects such as image colorization, super-resolution generation, and style transfer. At present, the existing image translation is to transform the problem into a one-to-one image mapping, which needs to be clearly given two different ima...

AI Technical Summary

Problems solved by technology

At present, the existing image translation is to transform the problem into a one-to-one image mapping, which needs to be clearly given two different image domains, and in many scenarios, cross-domain image translation is multimodal, so , the existing cross-domain data translation cannot meet these needs

Images