Towards Total Recall in Industrial Anomaly Detection - arXiv

On the challenging, widely used MVTec AD benchmark PatchCore achieves an image-level anomaly detection AUROC score of up to 99.6\%, more than ... AccessiblearXiv DoyounavigatearXivusingascreenreaderorotherassistivetechnology?Areyouaprofessorwhohelpsstudentsdoso?Wewanttohearfromyou. PleaseconsidersigninguptoshareyourinsightsasweworktomakearXivevenmoreopen. ShareInsights ComputerScience>ComputerVisionandPatternRecognition arXiv:2106.08265(cs) [Submittedon15Jun2021(v1),lastrevised5May2022(thisversion,v2)] Title:TowardsTotalRecallinIndustrialAnomalyDetection Authors:KarstenRoth,LathaPemula,JoaquinZepeda,BernhardSchölkopf,ThomasBrox,PeterGehler DownloadPDF Abstract:Beingabletospotdefectivepartsisacriticalcomponentinlarge-scale industrialmanufacturing.Aparticularchallengethatweaddressinthiswork isthecold-startproblem:fitamodelusingnominal(non-defective)example imagesonly.Whilehandcraftedsolutionsperclassarepossible,thegoalisto buildsystemsthatworkwellsimultaneouslyonmanydifferenttasks automatically.ThebestperformingapproachescombineembeddingsfromImageNet modelswithanoutlierdetectionmodel.Inthispaper,weextendonthisline ofworkandpropose\textbf{PatchCore},whichusesamaximallyrepresentative memorybankofnominalpatch-features.PatchCoreofferscompetitiveinference timeswhileachievingstate-of-the-artperformanceforbothdetectionand localization.Onthechallenging,widelyusedMVTecADbenchmarkPatchCore achievesanimage-levelanomalydetectionAUROCscoreofupto$99.6\%$,more thanhalvingtheerrorcomparedtothenextbestcompetitor.Wefurtherreport competitiveresultsontwoadditionaldatasetsandalsofindcompetitive resultsinthefewsamplesregime.\freefootnote{$^*$Workdoneduringa researchinternshipatAmazonAWS.}Code: thishttpURL. Comments: AcceptedtoCVPR2022 Subjects: ComputerVisionandPatternRecognition(cs.CV) Citeas: arXiv:2106.08265[cs.CV]   (or arXiv:2106.08265v2[cs.CV]forthisversion)   https://doi.org/10.48550/arXiv.2106.08265 Focustolearnmore arXiv-issuedDOIviaDataCite SubmissionhistoryFrom:KarstenRoth[viewemail] [v1] Tue,15Jun202116:27:02UTC(14,942KB)[v2] Thu,5May202223:01:53UTC(14,948KB) Full-textlinks: Download: PDF Otherformats (license) Currentbrowsecontext:cs.CV new | recent | 2106 Changetobrowseby: cs References&Citations NASAADSGoogleScholar SemanticScholar DBLP-CSBibliography listing|bibtex KarstenRothLathaPemulaJoaquinZepedaBernhardSchölkopfThomasBrox … a exportbibtexcitation Loading... Bibtexformattedcitation × loading... Dataprovidedby: Bookmark BibliographicTools BibliographicandCitationTools BibliographicExplorerToggle BibliographicExplorer(WhatistheExplorer?) LitmapsToggle Litmaps(WhatisLitmaps?) scite.aiToggle sciteSmartCitations(WhatareSmartCitations?) Code&Data CodeandDataAssociatedwiththisArticle arXivLinkstoCodeToggle arXivLinkstoCode&Data(WhatisLinkstoCode&Data?) Demos Demos ReplicateToggle Replicate(WhatisReplicate?) RelatedPapers RecommendersandSearchTools ConnectedPapersToggle ConnectedPapers(WhatisConnectedPapers?) Corerecommendertoggle CORERecommender(WhatisCORE?) AboutarXivLabs arXivLabs:experimentalprojectswithcommunitycollaborators arXivLabsisaframeworkthatallowscollaboratorstodevelopandsharenewarXivfeaturesdirectlyonourwebsite. BothindividualsandorganizationsthatworkwitharXivLabshaveembracedandacceptedourvaluesofopenness,community,excellence,anduserdataprivacy.arXiviscommittedtothesevaluesandonlyworkswithpartnersthatadheretothem. HaveanideaforaprojectthatwilladdvalueforarXiv'scommunity?LearnmoreaboutarXivLabsandhowtogetinvolved. Whichauthorsofthispaperareendorsers?| DisableMathJax(WhatisMathJax?)