AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Be Tangential to Manifold: Discovering Riemannian Metric for Diffusion Models

Shinnosuke Saito, Takashi Matsubara

ECCV 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
use score-Jacobian Riemannian metric geodesics instead of SLERPStable Diffusion v2.1-base · image interpolation (MB, CelebA-HQ, AFHQ)PPL: e.g. 0.871 -> 0.767 on AFHQImproved
use score-Jacobian metric instead of density-based GeoDiffStable Diffusion v2.1-base · image interpolationFID: 25.80 -> 21.01 on AFHQImproved
use score-Jacobian metric instead of FIM-based metricStable Diffusion v2.1-base · image interpolationPPL: 5.249 -> 0.767 on AFHQImproved
use metric instead of SLERP for PDVStable Diffusion v2.1-base · image interpolationPDV: best on MB(A) 0.021; SLERP better on AFHQ 0.022 vs 0.023Mixed
use metric geodesics for video frame interpolationStable Diffusion v2.1-base · video frame interpolationMSE: 13.253 -> 8.777 on DAVIS (vs GeoDiff)Improved
use LERP vs metric on DAVIS MSELERP · video frame interpolation DAVISMSE: 12.135 -> 8.777Improved
add metric-based guidance correction to CFGStable Diffusion v2.1-base · MS-COCO 30kFID: 17.28 -> 16.04 at w=12.5Improved
add metric-based guidance correction to CFGStable Diffusion v2.1-base · MS-COCO 30kCLIP Score: 0.315 -> 0.315 at w=12.5No clear effect
replace CFG with CFG++Stable Diffusion v2.1-base · MS-COCO 30kFID: 14.29 -> 13.98 at w=7.5; 11.69 -> 11.87 at w=5.0Mixed
apply guidance correction during LCM distillationLCM student · MS-COCO 4-stepFID: 17.91 -> 16.98Improved

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