AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

$$\Delta $$-DiT: Accelerating Diffusion Transformers without Training via Denoising Property Alignment

Pengtao Chen, Mingzhu Shen, P.D. Ye, Jianjian Cao, Chongjun Tu, Christos-Savvas Bouganis, Yiren Zhao, Tao Chen

IJCV 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration · 195 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
cache back blocks instead of front blocks (outline quality) (ablation)Delta-Cache · PIXART-alpha, 500 MS-COCO2017Average gradient: 36.170 -> 59.320Improved
cache front blocks instead of back blocks (detail quality) (ablation)Delta-Cache · PIXART-alpha, 500 MS-COCO2017PIQE: 19.974 -> 13.137Improved
apply Delta-Cache to back blocks early and front blocks late (stage-adaptive)Delta-Cache · MS-COCO2017FID: 34.819 -> 35.882Mixed
stage-adaptive Delta-DiT vs PIXART-alpha at same cost (T=13)PIXART-alpha · MS-COCO2017FID: 39.989 -> 35.882Improved
Delta-DiT vs PIXART-alpha speedupPIXART-alpha · MS-COCO2017latency (ms): 2290.668 -> 1534.551Improved
Delta-DiT vs DiT-XL (T=20)DiT-XL · ImageNet-50kFID: 15.893 -> 13.289Improved
Delta-DiT vs PIXART-alpha-LCMPIXART-alpha-LCM · MS-COCO2017FID: 40.433 -> 39.967Improved
use Delta-DiT with EulerD solver (ablation)PIXART-alpha · MS-COCO2017FID: 39.688 -> 35.735Improved
use Delta-DiT with DEIS solver (ablation)PIXART-alpha · MS-COCO2017FID: 37.675 -> 35.302Improved
reverse cache order (front early, back late) (ablation)Delta-DiT · MS-COCO2017FID, IS, CLIPHurt
increase boundary b from 0 to 20 (ablation)Delta-DiT · MS-COCO2017FID, IS, CLIPMixed
increase cached blocks Nc from 0 to 28 (ablation)Delta-DiT · MS-COCO2017FID, IS, CLIPMixed

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