AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Remove the Ambiguity: Few-shot Multimodal Anomaly Detection Using Crossmodal Feature Replacers

Yuan Guo, Wanqi Zhang, Xu Wang

ICML 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Unified multi-class anomaly detection

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 attention-aided retrieval and feature replacement (ablation)CFR · Eyecandies and MVTec 3D-AD one-to-many classes, 1-shot (mean)I-AUROC: 75.1 -> 81.3Improved
add attention-aided retrieval and feature replacement (ablation)CFR · Eyecandies and MVTec 3D-AD one-to-many classes, 1-shot (mean)30% AUPRO: 82.6 -> 85.4Improved
add attention-aided retrieval on Chocolate Praline (ablation)CFR · Eyecandies Chocolate Praline, 1-shotI-AUROC: 78.1 -> 86.7Improved
add attention-aided retrieval on Marshmallow (ablation)CFR · Eyecandies Marshmallow, 1-shot30% AUPRO: 91.2 -> 90.8Hurt
add cyclic reconstruction loss (CFR without retrieval vs CFM) (ablation)crossmodal reconstruction · 1-shotI-AUROC / 30% AUPRO: CFR without retrieval still outperforms CFMImproved
more training shots for crossmodal mapping (ablation)CFR · Eyecandies Hazelnut TruffleI-AUROC: 83.5 (1-shot) -> 71.5 (4-shot)Hurt
more samples used only in attention-aided retrieval, mapping shots fixed (ablation)CFR · Eyecandies Hazelnut TruffleI-AUROC: 81.8 -> 82.7 (1-shot CFL)Improved
vary replacement threshold theta from 0.0 to 0.5 (ablation)CFR · Licorice Sandwich, Marshmallow, Dowel, Peach 4-shotI-AUROC: 0.773 -> 0.854 (Licorice, theta 0.0 vs 0.2)No clear effect
CFR over reconstruction baseline at low-FPR AUPRO (ablation)CFR · Eyecandies 1-shot10% AUPRO: 61.5 -> 65.0Improved
CFR over M3DM on 1-shot MVTec 3D-AD image-level detectionCFR · MVTec 3D-AD 1-shot (mean)I-AUROC: 73.9 -> 74.0No clear effect

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