AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Foresight Diffusion: Improving Sampling Consistency in Predictive Diffusion Models

Yu Zhang, Xingzhuo Guo, Haoran Xu, Jialong Wu, Mingsheng Long

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space · 2 citations (OpenAlex)

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
add two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation)ForeDiff · RoboNetSTD of LPIPS: 0.66 -> 0.35Improved
add two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation)ForeDiff · RT-1STD of LPIPS: 0.50 -> 0.17Improved
add two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation)ForeDiff · HeterNSrelative L2: 0.83 -> 0.18Improved
add architectural decoupling alone (ForeDiff-zero) to vanilla diffusion (ablation)Vanilla Diffusion · RoboNetLPIPS (x100): 5.65 -> 5.54No clear effect
add architectural decoupling alone (ForeDiff-zero) to vanilla diffusion (ablation)Vanilla Diffusion · HeterNSrelative L2: 1.50 -> 0.83Improved
add predictive ViT stream and pretraining (full ForeDiff) to vanilla diffusion (ablation)Vanilla Diffusion · RoboNetFVD: 53.8 -> 51.5Improved
add predictive ViT stream and pretraining (full ForeDiff) to vanilla diffusion (ablation)Vanilla Diffusion · RT-1FVD: 11.7 -> 12.0Hurt
condition on internal predictive features instead of PredHead outputs (ablation)ForeDiff · RoboNetFVD: 53.7 -> 51.5Improved
increase predictive ViT blocks from M=0 to M=6 (ablation)ForeDiff · RoboNetLPIPS (x100): 5.65 -> 5.25Improved
increase predictive ViT blocks from M=6 to M=12 (ablation)ForeDiff · RoboNetFVD: 51.5 -> 52.1Hurt
add classifier-free guidance at inference (ablation)Vanilla Diffusion · RoboNetLPIPS (x100): 5.65 -> 5.27Improved
add classifier-free guidance at inference (ablation)ForeDiff · RoboNetLPIPS (x100): 5.25 -> 5.05Improved

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