AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Transition Models: Rethinking the Generative Learning Objective

Zidong Wang, Yiyuan Zhang, Xiaoyu Yue, Xiangyu Yue, Yangguang Li, Wanli Ouyang, Lei Bai

CVPR 2026 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space

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 SiT objective with TiM transition objective (ablation)SiT-B/4 · ImageNet-256FID (NFE=1): 309.5 -> 49.91Improved
replace SiT objective with TiM transition objective (ablation)SiT-B/4 · ImageNet-256FID (NFE=50): 20.35 -> 17.99Improved
use finite-difference DDE instead of JVP (ablation)TiM-B/4 · ImageNet-256FID (NFE=1): 49.75 -> 49.91No clear effect
use finite-difference DDE instead of JVP (ablation)TiM-B/4 · training costthroughput: 1.80 -> 2.40 (/s)Improved
add decoupled time embedding (ablation)TiM-B/4 · ImageNet-256FID (NFE=1): 56.22 -> 49.91Improved
add interval-aware attention (ablation)TiM-B/4 · ImageNet-256FID (NFE=1): 56.22 -> 48.38Improved
combine decoupled time embedding and interval-aware attention (ablation)TiM-B/4 · ImageNet-256FID (NFE=50): 20.37 -> 17.43Improved
add interval-aware attention on top of decoupled time embedding (ablation)TiM-B/4 · ImageNet-256FID (NFE=8): 26.09 -> 26.10No clear effect
add time-weighting favoring short intervals (ablation)TiM-B/4 · ImageNet-256FID (NFE=1): 48.30 -> 47.46Improved
increase sampling NFE from 1 to 128 (ablation)TiM · GenEvaloverall score: 0.67 -> 0.83Improved
increase sampling NFE from 1 to 128 (ablation)FLUX.1-Schnell · GenEvaloverall score: 0.68 -> 0.58Hurt
increase NFE from 8 to 32 (ablation)TiM · MJHQ30KFID: 5.28 -> 5.65Mixed

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