资源描述:
《neural algorithm of artistic style 》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ANeuralAlgorithmofArtisticStyleLeonA.Gatys,1;2;3AlexanderS.Ecker,1;2;4;5MatthiasBethge1;2;41WernerReichardtCentreforIntegrativeNeuroscienceandInstituteofTheoreticalPhysics,UniversityofTubingen,Germany¨2BernsteinCenterforComputationalNeuroscience,Tubingen,Germ
2、any¨3GraduateSchoolforNeuralInformationProcessing,Tubingen,Germany¨4MaxPlanckInstituteforBiologicalCybernetics,Tubingen,Germany¨5DepartmentofNeuroscience,BaylorCollegeofMedicine,Houston,TX,USATowhomcorrespondenceshouldbeaddressed;E-mail:leon.gatys@bethgelab.o
3、rgInfineart,especiallypainting,humanshavemasteredtheskilltocreateuniquevisualexperiencesthroughcomposingacomplexinterplaybetweenthecon-tentandstyleofanimage.Thusfarthealgorithmicbasisofthisprocessisunknownandthereexistsnoartificialsystemwithsimilarcapabilities.H
4、ow-ever,inotherkeyareasofvisualperceptionsuchasobjectandfacerecognitionnear-humanperformancewasrecentlydemonstratedbyaclassofbiologicallyinspiredvisionmodelscalledDeepNeuralNetworks.1,2HereweintroduceanarXiv:1508.06576v2[cs.CV]2Sep2015artificialsystembasedonaDe
5、epNeuralNetworkthatcreatesartisticimagesofhighperceptualquality.Thesystemusesneuralrepresentationstosepa-rateandrecombinecontentandstyleofarbitraryimages,providinganeuralalgorithmforthecreationofartisticimages.Moreover,inlightofthestrik-ingsimilaritiesbetweenp
6、erformance-optimisedartificialneuralnetworksandbiologicalvision,3–7ourworkoffersapathforwardtoanalgorithmicunder-standingofhowhumanscreateandperceiveartisticimagery.1TheclassofDeepNeuralNetworksthataremostpowerfulinimageprocessingtasksarecalledConvolutionalNeur
7、alNetworks.ConvolutionalNeuralNetworksconsistoflayersofsmallcomputationalunitsthatprocessvisualinformationhierarchicallyinafeed-forwardman-ner(Fig1).Eachlayerofunitscanbeunderstoodasacollectionofimagefilters,eachofwhichextractsacertainfeaturefromtheinputimage.T
8、hus,theoutputofagivenlayerconsistsofso-calledfeaturemaps:differentlyfilteredversionsoftheinputimage.WhenConvolutionalNeuralNetworksaretrainedonobjectrecognition,theydeveloparepresen