AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

FaPS: A General and Fast Training Method for Diffusion Models

Xianglu Wang, Bangxian Han, Hu Ding

ICML 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration

Read the paperCheck your idea against 137,217 papers

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
replace random patch selection with frequency-aware learned patch selectionPatch Diffusion · AFHQv2 CatFID: 3.11 vs. 2.98Improved
replace random patch selection with frequency-aware learned patch selectionPatch Diffusion · AFHQv2 DogFID: 4.80 vs. 4.73Improved
replace random patch selection with frequency-aware learned patch selectionPatch Diffusion · AFHQv2 WildFID: 1.93 vs. 1.63Improved
train with FaPS instead of full-image EDM-DDPM++EDM-DDPM++ · CelebA 64x64FID: 1.65 vs. 1.66No clear effect
train with FaPS instead of full-image EDM-DDPM++EDM-DDPM++ · FFHQ 64x64FID: 2.98 vs. 2.60Hurt
train with FaPS instead of LDM-ADMLDM-ADM · LSUN BedroomFID: 4.32 to 2.72Improved
train with FaPS instead of LDM-ADMLDM-ADM · LSUN ChurchFID: 4.66 to 2.73Improved
compare FaPS with LPDM-ADMLPDM-ADM · LSUN ChurchFID: 2.73 vs. 2.66Mixed
combine FaPS with SpeeD timestep samplingSpeeD · FFHQ 256x256 (DiT-XL/2)speedup: 1.0× -> 1.4×Improved
combine FaPS with SpeeD timestep samplingSpeeD · FFHQ 256x256 (DiT-XL/2)FID: 9.96 -> 10.85Hurt
add FaPS patch selection to DiT-XL/2 via MDTDiT-XL/2 · MetFaces 256x256FID: 29.34 -> 21.88Improved
use FaPS instead of MDT random maskingMDT · MetFaces 256x256speedup: 3.4× -> 3.8×Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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