Classifier-Free Guidance Diffusion Models for Image Synthesis
Автор: AI Focus
Загружено: 2025-09-29
Просмотров: 60
Classifier-free guidance jointly trains a conditional and an unconditional diffusion model without using a classifier, and then the resulting conditional and unconditional scores are combined to attain a trade-off between sample quality and diversity by selecting strength level similar to classifier guidance.
This video introduces how Classifier-free guidance defines a conditional and an unconditional model and combine their score estimates to tradeoff the FID and IS scores for sample quality and diversity.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: