AI Paper Atlas

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

DSR – A Dual Subspace Re-Projection Network for Surface Anomaly Detection

Vitjan Zavrtanik, Matej Kristan, Danijel Skočaj

ECCV 2022 · Label-efficient and robust learning › OOD and anomaly detection · approach: Unified multi-class anomaly detection · 201 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
generate anomalies in quantized feature space instead of image space (ablation)DSR · MVTecAP_loc: 67.5 -> 70.2Improved
generate anomalies in quantized feature space instead of image space (ablation)DSR · MVTecAUROC_det: 97.8 -> 98.2Improved
add similarity-bounded codebook sampling (ablation)DSR · MVTecAUROC_det: 95.6 -> 98.2Improved
add similarity-bounded codebook sampling (ablation)DSR · MVTecAP_loc: 62.5 -> 70.2Improved
use both image and feature reconstruction losses (ablation)DSR · MVTecAUROC_det: 96.3 -> 98.2Improved
use both image and feature reconstruction losses (vs feature only) (ablation)DSR · MVTecAP_loc: 61.8 -> 70.2Improved
add the upsampling module (ablation)DSR · MVTecAP_loc: 65.2 -> 70.2Improved
add the upsampling module (ablation)DSR · MVTecAUROC_det: 98.2 -> 98.2No clear effect
vary similarity bound lambda_s from 0.05 to 0.5 (ablation)DSR · MVTecAUROC_det: 98.2 -> 95.5Hurt
use a few segmented anomalous training images (N=16 instead of 0)DSR · KSDD2AP_det: 87.2 -> 91.4Improved
use full 246 segmented anomalous images instead of noneDSR · KSDD2AP_loc: 61.4 -> 85.5Improved
DSR vs fully supervised baseline at N=246DSR · KSDD2AP_det: 95.4 -> 95.2No 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