Consistent Diffusion Models Handle Noisy Data With Tweedie Consistency
Researchers propose Consistent Diffusion, a new approach to train diffusion models on noisy data using Tweedie consistency. This method outperforms existing methods on image generation tasks, especially with noisy input.
This is a Plain English Papers summary of a research paper called Robust diffusion models trained on noisy data with Tweedie consistency. If you like these kinds of analysis, you should join AImodels.fyi or follow me on Twitter. Overview This paper proposes a new diffusion model training approach called "Consistent Diffusion" that can handle noisy data and produce exact ambient diffusion models. The authors leverage the Tweedie distribution to model the noise in the data, which allows them to derive an exact training objective for the diffusion model. The resulting diffusion models a...