Generative de-noising device training and controllable generation method based on diffusion model distillation

By optimizing the diffusion model distillation training generative denoiser, combined with the teacher model, score model and discriminator, the problems of low sampling efficiency and insufficient controllability of the diffusion model are solved, and efficient and controllable image generation is achieved.

CN120471796APending Publication Date: 2025-08-12SHANGHAI JIAOTONG UNIV
0 Cites 4 Cited by

Patent Information

Application Number
CN202510567299.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-30
Publication Date
2025-08-12

AI Technical Summary

Technical Problem

The existing diffusion model has low sampling efficiency in image generation and lacks fine control and controllability of the generation process, resulting in insufficient quality and diversity of the generated results.

Method used

By using the pre-trained diffusion model as a teacher model, combining the generative denoiser, score model and discriminator, the training process of the generative denoiser is optimized, and the teacher diffusion model is used to estimate the data score and the scoring model is estimated, and the gradient is calculated and optimized until converges, and a trained generative denoiser is obtained.

Benefits of technology

It realizes the generation of high-quality samples in one-step sampling, and further improves sample quality through multi-step sampling, supports plug-and-play, and provides flexible and controllable sampling capabilities for various downstream tasks, significantly improving the generation efficiency and sample quality.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a training method of a generative denoising device and an image controllable generation and restoration method. The training method comprises the following steps: taking a pre-trained diffusion model as a teacher diffusion model; initializing a generative denoising device, a score model and a discriminator; obtaining a noiseless signal, and obtaining a noisy signal and Gaussian noise in combination with a generative denoising device; obtaining effective noise intensity and effective noise signals according to the noisy signals and the Gaussian noise; estimating a data score by using a teacher diffusion model according to the effective noise intensity, the effective noise signal and the noisy signal; according to the noisy signal, estimating a model score by using a score model; calculating and optimizing the gradient of learnable parameters of the generative denoising device; and according to the noisy signal, the data score, the model score and the optimized generative de-noising device, calculating de-noising score matching of the score model and adversarial loss of the discriminator, and optimizing the score model and the discriminator. And realizing image controllable generation and image restoration based on the generative de-noising device.
Need to check novelty before this filing date? Find Prior Art