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

Diffusion models are emerging expressive generative models, in which a largenumber of time steps (inference steps) are required for a single imagegeneration. To accelerate such tedious process, reducing steps uniformly isconsidered as an undisputed principle of diffusion models. We consider thatsuch a uniform assumption is not the optimal solution in practice; i.e., we canfind different optimal time steps for different models. Therefore, we proposeto search the optimal time steps sequence and compressed model architecture ina unified framework to achieve effective image generation for diffusion modelswithout any further training. Specifically, we first design a unified searchspace that consists of all possible time steps and various architectures. Then,a two stage evolutionary algorithm is introduced to find the optimal solutionin the designed search space. To further accelerate the search process, weemploy FID score between generated and real samples to estimate the performanceof the sampled examples. As a result, the proposed method is (i).training-free,obtaining the optimal time steps and model architecture without any trainingprocess; (ii). orthogonal to most advanced diffusion samplers and can beintegrated to gain better sample quality. (iii). generalized, where thesearched time steps and architectures can be directly applied on differentdiffusion models with the same guidance scale. Experimental results show thatour method achieves excellent performance by using only a few time steps, e.g.17.86 FID score on ImageNet 64 $\times$ 64 with only four steps, compared to138.66 with DDIM. The code is available athttps://github.com/lilijiangg/AutoDiffusion.

AutoDiffusion: Training-Free Optimization of Time Steps and Architectures for Automated Diffusion Model Acceleration