Training method of image generation model and image processing method and device

CN120913005APending Publication Date: 2025-11-07GUANGZHOU HUYA INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510939051.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-08
Publication Date
2025-11-07

AI Technical Summary

Technical Problem

In lightweight deep generative network models, the generator's generative ability cannot match the discriminator's discriminative ability, resulting in low image processing quality.

Method used

The quality of the generator's generated image is judged by the quality score of the generated image. When the quality score of the generated image is less than the first preset score threshold but greater than the second score threshold, a residual block is inserted into the discriminator, the depth of the discriminator is adjusted, and the discriminator after inserting the residual block is subjected to adversarial training with the generator until the quality score of the generator's generated image is less than the second score threshold. Then the depth of the discriminator is frozen, so as to realize the dynamic structural adaptation between the generator and the discriminator.

Benefits of technology

It improves the image generation quality of the generator, adapts the structure of the generator and discriminator, enhances image processing quality, and facilitates deployment on terminals.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120913005A_ABST
    Figure CN120913005A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides an image generation model training method and device and an image processing method and device, and relates to the technical field of image processing.The method comprises the steps that the generation quality of a generator is judged according to the generated image quality score of the generator, and when the generated image quality score is smaller than a first preset score threshold value and larger than a second score threshold value, image processing is conducted. And inserting residual blocks into the discriminator, performing adversarial training on the discriminator and the generator after the residual blocks are inserted, improving the image generation quality of the generator, and when the score of the generated image quality of the generator is smaller than a second score threshold, determining that the generation quality of the generator is relatively high and the number of the residual blocks in the discriminator is kept unchanged. According to the method, the depth of the discriminator is frozen, the training model of the discriminator with the number of residual blocks kept unchanged is trained to be converged, and the image generation model is obtained, so that the dynamic adaptation of the structure between the generator and the discriminator in the image generation model is realized, the image processing quality of the generator is also improved, and the generator can be conveniently deployed on a terminal.
Need to check novelty before this filing date? Find Prior Art