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 Probabilistic Models (DPM) have shown remarkable efficacy in thesynthesis of high-quality images. However, their inference processcharacteristically requires numerous, potentially hundreds, of iterative steps,which could exaggerate the problem of exposure bias due to the training andinference discrepancy. Previous work has attempted to mitigate this issue byperturbing inputs during training, which consequently mandates the retrainingof the DPM. In this work, we conduct a systematic study of exposure bias in DPMand, intriguingly, we find that the exposure bias could be alleviated with anovel sampling method that we propose, without retraining the model. Weempirically and theoretically show that, during inference, for each backwardtime step $t$ and corresponding state $\hat{x}_t$, there might exist anothertime step $t_s$ which exhibits superior coupling with $\hat{x}_t$. Based onthis finding, we introduce a sampling method named Time-Shift Sampler. Ourframework can be seamlessly integrated to existing sampling algorithms, such asDDPM, DDIM and other high-order solvers, inducing merely minimal additionalcomputations. Experimental results show our method brings significant andconsistent improvements in FID scores on different datasets and samplingmethods. For example, integrating Time-Shift Sampler to F-PNDM yields aFID=3.88, achieving 44.49\% improvements as compared to F-PNDM, on CIFAR-10with 10 sampling steps, which is more performant than the vanilla DDIM with 100sampling steps. Our code is available at https://github.com/Mingxiao-Li/TS-DPM.

Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps