Image generation method and model training method

CN115272788BActive Publication Date: 2026-07-14ALIBABA (CHINA) CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ALIBABA (CHINA) CO LTD
Filing Date
2022-07-14
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing image generation methods are prone to catastrophic loss of some information during the optimization process, resulting in low accuracy of image generation.

Method used

By acquiring the target vector and processing it using an image generation network, a target image is generated. The network parameters of the image generation network are updated based on the original results of the original image and the historical results of historical images, avoiding catastrophic forgetting and reusing historical information to improve the accuracy of image generation.

Benefits of technology

It improves the accuracy of image generation networks, enhances the performance of model reverse attack, avoids information loss during the optimization process, and improves the accuracy of image generation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115272788B_ABST
    Figure CN115272788B_ABST
Patent Text Reader

Abstract

The application discloses an image generation method and a model training method. The method comprises the following steps: obtaining a target vector, wherein the target vector satisfies a normal distribution; processing the target vector by using an image generation network to generate a target image, wherein the target image is used to represent a training sample of a target model, network parameters of the image generation network are updated based on an original result of an original image and a historical result of a historical image, the original result is obtained by processing the original image by using the target model, the historical result is obtained by processing the historical image by using the target model, the original image and the historical image are both generated by using the image generation network, the historical image is generated earlier than the original image, and the target model is a neural network model. The application solves the technical problem of low accuracy of an image generation method in the related art.
Need to check novelty before this filing date? Find Prior Art