Pixel-Aware Accelerated Reverse Diffusion Modeling
Keywords: diffusion models, generative models, accelerated learning, deep learning
TL;DR: An accelerated diffusion model based on exponentially decaying SNR in forward diffusion along with a parallel reverse trajectory generation.
Abstract: We propose in this paper an analytically new construct of a diffusion model whose drift and diffusion parameters yield a faster time-decaying Signal to Noise Ratio in the forward process. The proposed methodology significantly accelerates the forward diffusion process, reducing the required diffusion time steps from around 1000 seen in conventional models to 200-500 without compromising image quality in the reverse-time diffusion. In a departure from conventional models which typically use time-consuming multiple runs, we introduce a parallel data-driven model to generate a reverse-time diffusion trajectory in a single run of the model. The construct cleverly carries out the learning of the diffusion coefficients via an estimate of the structure of clean images. The resulting collective block-sequential generative model eliminates the need for MCMC-based sub-sampling correction for safeguarding and improving image quality, which further improve the acceleration of image generation. 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.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7507
Loading