English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

A diffusion model, which is formulated to produce an image using thousands ofdenoising steps, usually suffers from a slow inference speed. Existingacceleration algorithms simplify the sampling by skipping most steps yetexhibit considerable performance degradation. By viewing the generation ofdiffusion models as a discretized integrating process, we argue that thequality drop is partly caused by applying an inaccurate integral direction to atimestep interval. To rectify this issue, we propose a timestep aligner thathelps find a more accurate integral direction for a particular interval at theminimum cost. Specifically, at each denoising step, we replace the originalparameterization by conditioning the network on a new timestep, which isobtained by aligning the sampling distribution to the real distribution.Extensive experiments show that our plug-in design can be trained efficientlyand boost the inference performance of various state-of-the-art accelerationmethods, especially when there are few denoising steps. For example, when using10 denoising steps on the popular LSUN Bedroom dataset, we improve the FID ofDDIM from 9.65 to 6.07, simply by adopting our method for a more appropriateset of timesteps. Code will be made publicly available.

Towards More Accurate Diffusion Model Acceleration with A Timestep Aligner