Quick, Draw! Dataset - GitHub

The Quick Draw Dataset is a collection of 50 million drawings across 345 categories, contributed by players of the game Quick, Draw! Skiptocontent {{message}} googlecreativelab / quickdraw-dataset Public Notifications Fork 734 Star 4.8k DocumentationonhowtoaccessandusetheQuick,Draw!Dataset. quickdraw.withgoogle.com/data Viewlicense 4.8k stars 734 forks Star Notifications Code Issues 24 Pullrequests 2 Actions Wiki Security Insights More Code Issues Pullrequests Actions Wiki Security Insights Thiscommitdoesnotbelongtoanybranchonthisrepository,andmaybelongtoaforkoutsideoftherepository. master Branches Tags Couldnotloadbranches Nothingtoshow {{refName}} default Couldnotloadtags Nothingtoshow {{refName}} default 2 branches 0 tags Code Latestcommit HalfdanJ Updatelinkstogooglecloud … 641379b Feb7,2022 Updatelinkstogooglecloud 641379b Gitstats 64 commits Files Permalink Failedtoloadlatestcommitinformation. Type Name Latestcommitmessage Committime examples n_strokes&n_pointsisuint16 May28,2020 LICENSE Addinitialreadmeandexample May18,2017 README.md Updatelinkstogooglecloud Feb7,2022 categories.txt Addinitialreadmeandexample May18,2017 preview.jpg Addinitialreadmeandexample May18,2017 Viewcode TheQuick,Draw!Dataset Content Therawmoderateddataset Preprocesseddataset SimplifiedDrawingfiles(.ndjson) Binaryfiles(.bin) Numpybitmaps(.npy) Getthedata Fulldatasetseperatedbycategories Sketch-RNNQuickDrawDataset Projectsusingthedataset Changes License DatasetMetadata README.md TheQuick,Draw!Dataset TheQuickDrawDatasetisacollectionof50milliondrawingsacross345categories,contributedbyplayersofthegameQuick,Draw!.Thedrawingswerecapturedastimestampedvectors,taggedwithmetadataincludingwhattheplayerwasaskedtodrawandinwhichcountrytheplayerwaslocated.Youcanbrowsetherecognizeddrawingsonquickdraw.withgoogle.com/data. We'resharingthemherefordevelopers,researchers,andartiststoexplore,study,andlearnfrom.Ifyoucreatesomethingwiththisdataset,pleaseletusknowbye-mailoratA.I.Experiments. Wehavealsoreleasedatutorialandmodelfortrainingyourowndrawingclassifierontensorflow.org. Pleasekeepinmindthatwhilethiscollectionofdrawingswasindividuallymoderated,itmaystillcontaininappropriatecontent. Content Therawmoderateddataset Preprocesseddataset Getthedata Projectsusingthedataset Changes License Therawmoderateddataset Therawdataisavailableasndjsonfilesseperatedbycategory,inthefollowingformat: Key Type Description key_id 64-bitunsignedinteger Auniqueidentifieracrossalldrawings. word string Categorytheplayerwaspromptedtodraw. recognized boolean Whetherthewordwasrecognizedbythegame. timestamp datetime Whenthedrawingwascreated. countrycode string Atwolettercountrycode(ISO3166-1alpha-2)ofwheretheplayerwaslocated. drawing string AJSONarrayrepresentingthevectordrawing Eachlinecontainsonedrawing.Here'sanexampleofasingledrawing:  { "key_id":"5891796615823360", "word":"nose", "countrycode":"AE", "timestamp":"2017-03-0120:41:36.70725UTC", "recognized":true, "drawing":[[[129,128,129,129,130,130,131,132,132,133,133,133,133,...]]] } Theformatofthedrawingarrayisasfollowing: [ [//Firststroke [x0,x1,x2,x3,...], [y0,y1,y2,y3,...], [t0,t1,t2,t3,...] ], [//Secondstroke [x0,x1,x2,x3,...], [y0,y1,y2,y3,...], [t0,t1,t2,t3,...] ], ...//Additionalstrokes ] Wherexandyarethepixelcoordinates,andtisthetimeinmillisecondssincethefirstpoint.xandyarereal-valuedwhiletisaninteger.Therawdrawingscanhavevastlydifferentboundingboxesandnumberofpointsduetothedifferentdevicesusedfordisplayandinput. Preprocesseddataset We'vepreprocessedandsplitthedatasetintodifferentfilesandformatstomakeitfasterandeasiertodownloadandexplore. SimplifiedDrawingfiles(.ndjson) We'vesimplifiedthevectors,removedthetiminginformation,andpositionedandscaledthedataintoa256x256region.Thedataisexportedinndjsonformatwiththesamemetadataastherawformat.Thesimplificationprocesswas: Alignthedrawingtothetop-leftcorner,tohaveminimumvaluesof0. Uniformlyscalethedrawing,tohaveamaximumvalueof255. Resampleallstrokeswitha1pixelspacing. SimplifyallstrokesusingtheRamer–Douglas–Peuckeralgorithmwithanepsilonvalueof2.0. Thereisanexampleinexamples/nodejs/simplified-parser.jsshowinghowtoreadndjsonfilesinNodeJS. Additionally,theexamples/nodejs/ndjson.mddocumentdetailsasetofcommand-linetoolsthatcanhelpexploresubsetsofthesequitelargefiles. Binaryfiles(.bin) Thesimplifieddrawingsandmetadataarealsoavailableinacustombinaryformatforefficientcompressionandloading. Thereisanexampleinexamples/binary_file_parser.pyshowinghowtoloadthebinaryfilesinPython. Thereisalsoanexampleinexamples/nodejs/binary-parser.jsshowinghowtoreadthebinaryfilesinNodeJS. Numpybitmaps(.npy) Allthesimplifieddrawingshavebeenrenderedintoa28x28grayscalebitmapinnumpy.npyformat.Thefilescanbeloadedwithnp.load().Theseimagesweregeneratedfromthesimplifieddata,butarealignedtothecenterofthedrawing'sboundingboxratherthanthetop-leftcorner.Seehereforcodesnippetusedforgeneration. Getthedata ThedatasetisavailableonGoogleCloudStorageasndjsonfilesseperatedbycategory.SeethelistoffilesinCloud ,orreadmoreaboutaccessingpublicdatasetsusingothermethods.Asanexample,toeasilydownloadallsimplifieddrawings,onewayistorunthecommandgsutil-mcp'gs://quickdraw_dataset/full/simplified/*.ndjson'. Fulldatasetseperatedbycategories Rawfiles(.ndjson) Simplifieddrawingsfiles(.ndjson) Binaryfiles(.bin) Numpybitmapfiles(.npy) Sketch-RNNQuickDrawDataset ThisdataisalsousedfortrainingtheSketch-RNNmodel.Anopensource,TensorFlowimplementationofthismodelisavailableintheMagentaProject,(linktoGitHubrepo).YoucanalsoreadmoreaboutthismodelinthisGoogleResearchblogpost.Thedataisstoredincompressed.npzfiles,inaformatsuitableforinputsintoarecurrentneuralnetwork. Inthisdataset,75Ksamples(70KTraining,2.5KValidation,2.5KTest)hasbeenrandomlyselectedfromeachcategory,processedwithRDPlinesimplificationwithanepsilonparameterof2.0.Eachcategorywillbestoredinitsown.npzfile,forexample,cat.npz. Wehavealsoprovidedthefulldataforeachcategory,ifyouwanttousemorethan70Ktrainingexamples.Thesearestoredwiththe.full.npzextensions. Numpy.npzfiles Note:ForPython3,loadingthenpzfilesusingnp.load(data_filepath,encoding='latin1',allow_pickle=True) InstructionsforconvertingRawndjsonfilestothisnpzformatisavailableinthisnotebook. Projectsusingthedataset Herearesomeprojectsandexperimentsthatareusingorfeaturingthedatasetininterestingways.Gotsomethingtoadd?Letusknow! Creativeandartisticprojects LettercollagesbyDeborahSchmidt FacetrackingexperimentbyNeilMendoza FacesofHumanitybyTortue InfiniteQuickDrawbykynd.info Misfire.iobyMatthewCollyer DrawThisbyDanMacnish ScribblingSpeechbyXinyueYang illustrAItionbyLingChen DreamingofElectricSheepby Dr.ErnestoDiaz-Aviles Dataanalyses Howdoyoudrawacircle?byQuartz FormaFluensbyMauroMartino,HendrikStrobeltandOwenCornec HowLongDoesitTaketo(Quick)DrawaDog?byJimVallandingham FindingbadflamingodrawingswithrecurrentneuralnetworksbyColinMorris FacetsDivexQuick,Draw!byPeople+AIResearchInitiative(PAIR),Google ExploringandVisualizinganOpenGlobalDatasetbyGoogleResearch MachineLearningforVisualization-Talk/articlebyIanJohnson Papers ANeuralRepresentationofSketchDrawingsbyDavidHa,DouglasEck,ICLR2018.code Sketchmate:Deephashingformillion-scalehumansketchretrievalbyPengXuetal.,CVPR2018. Multi-graphtransformerforfree-handsketchrecognitionbyPengXu,ChaitanyaKJoshi,XavierBresson,ArXiv2019.code DeepSelf-SupervisedRepresentationLearningforFree-HandSketchbyPengXuetal.,ArXiv2020.code SketchTransfer:AChallengingNewTaskforExploringDetail-InvarianceandtheAbstractionsLearnedbyDeepNetworksbyAlexLamb,SherjilOzair,VikasVerma,DavidHa,WACV2020. DeepLearningforFree-HandSketch:ASurveybyPengXu,ArXiv2020. ANovelSketchRecognitionModelbasedonConvolutionalNeuralNetworksbyAbdullahTalhaKabakus,2ndInternationalCongressonHuman-ComputerInteraction,OptimizationandRoboticApplications,pp.101-106,2020. Guides&Tutorials TensorFlowtutorialfordrawingclassification Trainamodelintf.keraswithColab,andrunitinthebrowserwithTensorFlow.jsbyZaidAlyafeai Codeandtools Quick,Draw!PolymerComponent&DataAPIbyNickJonas Quick,DrawforProcessingbyCodyBenLewis Quick,Draw!predictionmodelbyKeisukeIrie RandomsampletoolbyLearningstatisticsisawesome SVGrenderingind3.jsexamplebyIanJohnson(readmoreabouttheprocesshere) Sketch-RNNClassificationbyPayalBajaj quickdraw.jsbyThomasWagenaar ~Doodler~by KrishnaSriSomepalli quickdrawPythonAPIbyMartinO'Hanlon RealTimeQuickDrawbyAkshayBahadur DataFlowprocessingbyGuillemXercavins QuickDrawGHRhinoPluginbyJamesDalessandro Changes May25,2017:UpdatedSketch-RNNQuickDrawdataset,created.full.npzcomplementarysets. License ThisdatamadeavailablebyGoogle,Inc.undertheCreativeCommonsAttribution4.0Internationallicense. DatasetMetadata Thefollowingtableisnecessaryforthisdatasettobeindexedbysearch enginessuchasGoogleDatasetSearch. property value name TheQuick,Draw!Dataset alternateName QuickDrawDataset alternateName quickdraw-dataset url https://github.com/googlecreativelab/quickdraw-dataset sameAs https://github.com/googlecreativelab/quickdraw-dataset description TheQuickDrawDatasetisacollectionof50milliondrawingsacross345categories,contributedbyplayersofthegame"Quick,Draw!".Thedrawingswerecapturedastimestampedvectors,taggedwithmetadataincludingwhattheplayerwasaskedtodrawandinwhichcountrytheplayerwaslocated.\n \n Exampledrawings:  provider property value name Google sameAs https://en.wikipedia.org/wiki/Google license property value name CCBY4.0 url https://creativecommons.org/licenses/by/4.0/ About DocumentationonhowtoaccessandusetheQuick,Draw!Dataset. quickdraw.withgoogle.com/data Topics dataset quickdraw-dataset Resources Readme License Viewlicense Stars 4.8k stars Watchers 194 watching Forks 734 forks Releases Noreleasespublished Packages0 Nopackagespublished Contributors20 +9contributors Youcan’tperformthatactionatthistime. Yousignedinwithanothertaborwindow.Reloadtorefreshyoursession. Yousignedoutinanothertaborwindow.Reloadtorefreshyoursession.