AI Paper Atlas

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

Multi-Prototype Compactness and Boundary-Aware Synthesis for Unsupervised Anomaly Detection

Kailun Liao, Jianfeng Yang, Tao Tao, Wenfei Wu, Jiaming Jiang, Jinsheng Xiao

CVPR 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 multi-prototype compact constraint (MPCC) to single-prototype baseline (ablation)PGBL baseline · MVTec-ADI-AUROC: 96.3 -> 97.0Improved
add DBR discriminator trained on Gaussian-noise pseudo-anomalies (ablation)MPCC model · MVTec-ADI-AUROC: 97.0 -> 99.2Improved
replace Gaussian-noise anomaly synthesis with boundary-aware synthesis (BAAS) (ablation)PGBL · MVTec-ADI-AUROC: 99.2 -> 99.8Improved
replace Gaussian-noise anomaly synthesis with BAAS (ablation)PGBL · MVTec-ADP-AUPRO: 94.3 -> 95.0Improved
vary number of cluster prototypes M from 1 to 30 (ablation)PGBL · MVTec-AD and VisAI-AUROC: peak at M = 10: 99.8 (MVTec-AD), 98.0 (VisA)Mixed
vary k-NN neighborhood size k in BAAS (ablation)PGBL · MVTec-ADI-AUROCNo clear effect
increase noise variance sigma^2 to very large values (ablation)BAAS · MVTec-ADI-AUROCHurt
change pretrained backbone (ResNet50, ResNet101, WideResNet50, ViT-B/16, ViT-L/16) (ablation)PGBL · MVTec-ADI-AUROC / P-AUROC: WideResNet50 P-AUROC 98.6No clear effect
change random seed for prototype initialization (ablation)PGBL · MVTec-ADI-AUROC: 99.8 / 99.8 / 99.7 / 99.8No clear effect
use PGBL instead of SimpleNet (adds BAAS k-NN)inference/training cost · Real-IADtraining latency (ms): 7.83 -> 21.12Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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