Accelerated Image-Aware Diffusion Modeling

Tanmay Asthana, Yufang Bao, Amro Awad, Hamid Krim

Abstract: We propose in this paper an analytically new construct of a Diffusion Model whose drift and diffusion parameters yield an accelerated time-decaying Signal-to-Noise Ratio (SNR) in the forward process. This consequently reduces the number of time steps required to converge to pure noise. It further allows us to depart from conventional models, which typically use time-consuming multiple runs, by introducing a parallel datadriven model to generate a reverse-time diffusion trajectory in a single run. Our construct cleverly carries out the learning of the diffusion coefficients on the structure of clean images using an autoencoder. Collectively, these advancements yield a generative model that is at least 4 times faster than conventional approaches, while maintaining high fidelity and diversity in generated images, hence promising widespread applicability in rapid image synthesis tasks.

External IDs:dblp:conf/eusipco/AsthanaBAK25