搜索
牛津建筑物数据集(TheOxfordBuildingsDataset)数据介绍：TheOxfordBuildingsDatasetconsistsof5062imagescollectedfromFlickrbysearchingforparticularOxfordlandmarks.Thecollectionhasbeenmanuallyannotatedtogenerateacomprehensivegroundtruthfor11differentlandmarks,eachrepresentedby5possiblequeries.Thisgivesasetof55queriesoverwhichanobjectretrievalsystemcanbeevaluated.关键词：牛津建筑物,特别,标志性建筑,地面实况,检索,OxfordBuildings,particular,landmarks,groundtruth,retrieval,数据格式：IMAGE数据详细介绍：TheOxfordBuildingsDatasetJamesPhilbinandAndrewZissermanOverviewTheOxfordBuildingsDatasetconsistsof5062imagescollectedfromFlickrbysearchingforparticularOxfordlandmarks.Thecollectionhasbeenmanuallyannotatedtogenerateacomprehensivegroundtruthfor11differentlandmarks,eachrepresentedby5possiblequeries.Thisgivesasetof55queriesoverwhichanobjectretrievalsystemcanbeevaluated.GroundtruthQueriesThefollowingimageshowsall55queriesusedtoevaluateperformanceoverthegroundtruth.Flickrqueriesusedtodownloadimages1.AllSoulsOxford2.BalliolOxford3.ChristChurchOxford4.HertfordOxford5.JesusOxford6.KebleOxford7.MagdalenOxford8.NewOxford9.OrielOxford10.TrinityOxford11.RadcliffeCameraOxford12.CornmarketOxford13.BodleianOxford14.PittRiversOxford15.AshmoleanOxford16.WorcesterOxford17.OxfordForeachimageandlandmarkinourdataset,oneoffourpossiblelabelswasgenerated:1.Good-Anice,clearpictureoftheobject/building.2.OK-Morethan25%oftheobjectisclearlyvisible.3.Bad-Theobjectisnotpresent.4.Junk-Lessthan25%oftheobjectisvisible,orthereareveryhighlevelsofocclusionordistortion.DatabaseRightsTheOxfordbuildingsdatasetconsistsofimagesprovidedbyFlickr.Useoftheseimagesmustrespectthecorresponingtermsofuse.Theidentityoftheimagesinthedatabasehasbeenobscured.Anyqueriesabouttheuseorownershipofthedatashouldbeaddressedhere.DownloadsAlldataneededforevaluationisgivenbelow:1.Datasetimages2.Groundtruthfiles3.C++codetocomputethegroundtruthAdditionally,we'vemadeextradataavailableforthe5Kdataset:1.READMEforthefollowingfiles2.CompressedbinaryfileofSIFTdescriptorsforthe5Kdataset[MD5:fbc0e85c5065f6d97d519a7f2ed3e3f9]3.CompressedtextfilescontainingwordIDsandgeometryforthe5Kdatasetusingavocabularysizeof1M[MD5:0734e6023f6bf2140b9531af22ce7953]Computingtheaverageprecision1.Compilethecompute_ap.cppfile,using(onlinux)g++-Ocompute_ap.cpp-ocompute_ap.2.Tocomputetheaverageprecisionforarankedlist,rank_list.txt,oneruns./compute_apchrist_church_1rank_list.txt,forthefirstChristChurchquery,etc.RelevantPublicationPhilbin,J.,Chum,O.,Isard,M.,Sivic,J.andZisserman,A.ObjectretrievalwithlargevocabulariesandfastspatialmatchingProceedingsoftheIEEEConferenceonComputerVisionandPatternRecognition(2007)Bibtexsource|Abstract|Document:ps.gzPDFAcknowledgementsThisworkissupportedbyanEPSRCPlatformgrant.数据预览：点此下载完整数据集