AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

ScoreMix: Synthetic Data Generation by Score Composition in Diffusion Models Improves Recognition

Parsa Rahimi, Sébastien Marcel

ICML 2026 · Image generation › Fast diffusion sampling · approach: Diffusion models as perception tools · 0 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 ScoreMix synthetic augmentation to real dataIR50 ArcFace · IJB-C TAR@FPR 1e-6TAR: 70.37 -> 76.45Improved
add ScoreMix augmentation instead of scaling backbone to IR101IR50 ArcFace · IJB-B TAR@FPR 1e-6TAR: 34.84 -> 35.95Improved
use ScoreMix instead of AugGen augmentationIR50 ArcFace · IJB-C TAR@FPR 1e-6TAR: 75.02 -> 76.45Improved
mix most distant classes in embedding space rather than closest (ablation)ScoreMix · class selection averageAvg: 64.92 -> 67.44Improved
mix most distant classes in condition space (cosine) rather than closest (ablation)ScoreMix · class selection averageAvg: 66.84 -> 66.95No clear effect
mix most distant classes in condition space (L2) rather than closest (ablation)ScoreMix · class selection averageAvg: 65.23 -> 65.79No clear effect
choose embedding-distant pairs instead of random pairs (ablation)ScoreMix · class selection averageAvg: 66.17 -> 67.44Improved
mix three classes instead of two (ablation)ScoreMix · class selection averageAvg: 67.44 -> 65.69Hurt
select pairs by close condition and distant embedding (ablation)ScoreMix · class selection averageAvg: 67.44 -> 66.98No clear effect
select pairs using a different discriminator (AdaFace-IR101) instead of self (ablation)ArcFace-IR50 target · cross-architecture selectionPerf: 68.08 -> 67.62No clear effect
align generator outputs to class centers via regularization (ablation)generator reproduction · reproduction-set recognitionAvgHurt
add ScoreMix augmentation to real dataIR50 ArcFace · training time on WebFace160Khours: 2.54h -> 4.10hHurt

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