AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Reflected Diffusion Models

Aaron Lou, Stefano Ermon

ICML 2023 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space · 3 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use reflected diffusion instead of the strongest listed baselineCIFAR-10 image generation · CIFAR-10Inception Score: 10.11 -> 10.46Improved
Use reflected diffusion instead of the strongest listed baselineCIFAR-10 image generation · CIFAR-10FID: 1.97 -> 2.72Hurt
Use reflected diffusion without a modified noise schedule instead of VDM with a modified noise scheduleCIFAR-10 likelihood estimation · CIFAR-10Bits per dimension: 2.65 -> 2.68Hurt
Use reflected diffusion without a modified noise schedule instead of VDM with a modified noise scheduleImageNet32 likelihood estimation · ImageNet32Bits per dimension: 3.72 -> 3.74Hurt
Add predictor-corrector sampling with a small signal-to-noise ratio (ablation)Reflected Diffusion · CIFAR-10image generation qualityImproved
Use reflection rather than projection in the reflected Euler-Maruyama discretization (ablation)Reflected Diffusion sampling · image generationsample qualityImproved
Apply dynamic thresholding rather than reflected sampling (ablation)guided Reflected Diffusion · ImageNet64sample fidelityHurt
Remove thresholding from standard diffusion sampling (ablation)standard diffusion with classifier-free guidance · ImageNet64divergent sample rate: about halfHurt
Remove thresholding from standard diffusion sampling (ablation)standard diffusion with classifier guidance · ImageNet64divergent sample rate: around 75\%Hurt
Use reflected diffusion rather than thresholded standard diffusion at high guidance weightclass-conditioned image generation · ImageNet64, guidance weight w=15sample fidelityImproved
Use reflected ODE sampling for classifier-free guidanceReflected Diffusion · ImageNet64, guidance weight w=1.5forward evaluations: around 100 compared with 1000Improved
Use the denoising autoencoder to remove noise from reflected-diffusion samples (ablation)Reflected Diffusion · CIFAR-10FIDImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy