Face attribute editing method based on balance stack generative adversarial network

Abstract

The invention discloses a face attribute editing method based on a balance stack generative adversarial network, which comprises the following steps: 1) obtaining a data set containing a face image and an attribute label, and preprocessing the data set; 2) constructing a plurality of conditional generative adversarial networks composed of paired generators and discriminators according to the sizeof the face image; 3) independently training all conditional generative adversarial networks by utilizing the preprocessed data set and aiming at different face attributes and adopting a weighted learning and residual image generation mode; and 4) stacking all trained generators to form a stack structure, and sequentially editing corresponding face attributes for the preprocessed unknown face image. Weighted learning and residual image generation modes are applied in the training process, and a stack structure is adopted in the attribute editing process, so that the model can effectively copewith the problem of data imbalance, the editing capability for minority class samples is enhanced, the image generation effect in an attribute irrelevant region is improved, and the problem of attribute entanglement is avoided.

Description

technical field [0001] The present invention relates to the technical fields of image editing and machine learning, in particular to a face attribute editing method based on a balanced stack generative confrontation network. Background technique [0002] Face attribute editing focuses on changing specific attributes under a given face image, such as adding glasses, removing beard, whitening skin or even replacing gender. This vision task itself achieves the controllability of image semantics and fine-grained image transformation. At the same time, with the rise of the wave of selfies on social media, the widespread dissemination of online videos, and the intelligent design of game or animation characters, large-scale face image data is generated on the Internet every year, and there is an increasingly strong demand for face attribute editing. Therefore, face attribute editing is widely used in application scenarios such as beauty, video repair, and character synthesis. In ...

AI Technical Summary

Problems solved by technology

[0005] In order to achieve the above object, the technical solution provided by the present invention is: a face attribute editing method based on a balanced stack generative confrontation network, which uses weighted learning and training multiple conditional generative confrontation networks to solve The problem of attribute imbalance, and stacking all the generators of the trained conditional generative confrontation network to form a stack structure, solve the problem of attribute entanglement, and use the residual image generation method to solve the problem of inaccurate image editing; it includes the following steps:

Images