Editing model generation method and device, face image editing method and device, equipment and medium

Abstract

The embodiment of the invention discloses an editing model generation method and device, a human face image editing method and device, equipment and a medium. The editing model generation method comprises the steps: training a generative adversarial model comprising a generator and a discriminator; updating the generative adversarial model according to gradient updating configuration information of the discriminator, the gradient updating configuration information being determined through a Lipschitz constraint condition; and when determining that the generative adversarial model meets a training ending condition, generating an image editing model according to a generator in the currently trained generative adversarial model. According to the embodiment of the invention, the training consistency of the generator and the discriminator can be improved, and the authenticity of the generated image is improved.

Description

technical field [0001] The embodiments of the present invention relate to the field of artificial intelligence, and in particular to a method, device, equipment and medium for generating an editing model and editing a face image. Background technique [0002] In recent years, people have higher and higher requirements for the authenticity of synthetic images, which requires image processing algorithms to generate more realistic and natural images. In particular, people often edit face images, expecting that the edited face images are still real faces. [0003] At present, Generative Adversarial Network (GAN) can be used to generate real faces. Among them, in the training process of the generative confrontation model, the generator in the generative confrontation network is actually used to generate face images, and the discriminator in the generative confrontation model is used to distinguish the authenticity of the generated face images. [0004] The training of the gener...

AI Technical Summary

Problems solved by technology

When the discriminator quickly completes the training process in one step and can accurately judge the authenticity of the face image, the generator cannot pass the judgment of the discriminator no matter how much progress it makes, which makes the training of the generative confrontation network fail, and the generated face cannot be guaranteed. The authenticity of the image, and thus cannot guarantee the authenticity of the editing effect of the image editing model based on the pre-trained generation confrontation network structure for the face image

Images