A conditional guided image translation method based on a diffusion model

By combining a diffusion model with a conditional generative adversarial network, and using a residual attention network to extract deep feature maps and jointly train them with the diffusion model, the problem of unstable quality in infrared remote sensing image generation is solved, and efficient and high-fidelity image translation is achieved.

CN118822841BActive Publication Date: 2026-07-24HARBIN INST OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HARBIN INST OF TECH
Filing Date
2024-06-25
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing technologies struggle to generate high-quality infrared remote sensing images for specific categories, and the dependence of conditional generative adversarial networks on conditional information leads to unstable generation quality.

Method used

By combining a diffusion model with a conditional generative adversarial network, a deep global feature map is extracted as conditional information through a residual attention network. The feature map is then refined using a diffusion model and jointly trained with the conditional generative adversarial network to generate high-quality infrared remote sensing images.

Benefits of technology

It achieves fast, high-quality, and high-fidelity image translation in infrared remote sensing image generation, while reducing computational resource consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118822841B_ABST
    Figure CN118822841B_ABST
Patent Text Reader

Abstract

The application discloses a conditional guiding image translation method based on a diffusion model, and proposes a two-stage image translation model combining a diffusion model and a conditional generative adversarial network. In the first stage, ResAttNet1 is pre-trained to extract deep features, and rich semantic information contained in the deep feature map is used as conditional information of the conditional generative adversarial network to guide the conditional generative adversarial network to complete image translation. In the second stage, the trained ResAttNet1 and the parameter randomly initialized ResAttNet2 are used to extract global feature information and sample level deep features respectively, a strategy of jointly training the conditional generative adversarial network and the diffusion model is adopted, a light diffusion model is used to refine the deep features, and finally an image translation network with stable training, good generated image fidelity and high sampling rate is constructed. The method can improve the quality and accuracy of the conditional information and effectively improve the image translation performance of the CGAN.
Need to check novelty before this filing date? Find Prior Art