Architecture
Denoising Diffusion Probabilistic Models
UC Berkeley·June 19, 2020
Jonathan Ho, Ajay Jain, Pieter Abbeel
View on arXivTL;DR
Reformulates diffusion models and shows they can generate high-quality images, establishing the denoising framework used by modern image generators.
Why it matters
The technical foundation of the image-generation boom — nearly every modern text-to-image system descends from this denoising formulation.