learning to generate chairs with convolutional neural networks

learning to generate chairs with convolutional neural networks

ID：40720157

大小：5.01 MB

頁數(shù)：13頁

時間：2019-08-06

learning to generate chairs with convolutional neural networks_第1頁
learning to generate chairs with convolutional neural networks_第2頁
learning to generate chairs with convolutional neural networks_第3頁
learning to generate chairs with convolutional neural networks_第4頁
learning to generate chairs with convolutional neural networks_第5頁
資源描述：

《learning to generate chairs with convolutional neural networks》由會員上傳分享，免費在線閱讀，更多相關內(nèi)容在學術論文-天天文庫。

1、LearningtoGenerateChairswithConvolutionalNeuralNetworksAlexeyDosovitskiyJostTobiasSpringenbergThomasBroxDepartmentofComputerScience,UniversityofFreiburgfdosovits,springj,broxg@cs.uni-freiburg.deAbstractWetrainagenerativeconvolutionalneuralnetworkwhichisabletogenerateimagesofobject

2、sgivenobjecttype,viewpoint,andcolor.Wetrainthenetworkinasu-pervisedmanneronadatasetofrendered3Dchairmod-els.Ourexperimentsshowthatthenetworkdoesnotmerelylearnallimagesbyheart,butrather?ndsameaningfulrepresentationofa3Dchairmodelallowingittoassessthesimilarityofdifferentchairs,inte

3、rpolatebetweengivenviewpointstogeneratethemissingones,orinventnewchairstylesbyinterpolatingbetweenchairsfromthetrainingset.Figure1.Interpolationbetweentwochairmodels(original:topWeshowthatthenetworkcanbeusedto?ndcorrespon-left,?nal:bottomleft).Thegenerativeconvolutionalneuralnet-w

4、orklearnsthemanifoldofchairs,allowingittointerpolatebe-dencesbetweendifferentchairsfromthedataset,outper-tweenchairstyles,producingrealisticintermediatestyles.formingexistingapproachesonthistask.canperfectlyapproximateanyfunctiononthetrainingset.1.IntroductionInourcase,anetworkpot

5、entiallycouldjustlearnbyheartallexamplesandprovideperfectreconstructionsofthese,Convolutionalneuralnetworks(CNNs)havebeenshownbutwouldbehaveunpredictablywhenconfrontedwithin-tobeverysuccessfulonavarietyofcomputervisiontasks,putsithasnotseenduringtraining.Weshowthatthisisnotsuchasi

6、mageclassi?cation[17,5,31],detection[9,27]whatishappening,bothbecausethenetworkistoosmalltoandsegmentation[9].Allthesetaskshaveincommonjustrememberallimages,andbecauseweobservegener-thattheycanbeposedasdiscriminativesupervisedlearn-alizationtopreviouslyunseendata.Namely,weshowthat

7、ingproblems,andhencecanbesolvedusingCNNswhichthenetworkiscapableof:1)knowledgetransfer:givenlim-areknowntoperformwellgivenalargeenoughlabeleditednumberofviewpointsofanobject,thenetworkcanusedataset.Typically,atasksolvedbysupervisedCNNsin-theknowledgelearnedfromothersimilarobjectst

8、oinfervolveslearningmappingsfromr

當前文檔最多預覽五頁，下載文檔查看全文

此文檔下載收益歸作者所有

當前文檔最多預覽五頁，下載文檔查看全文
溫馨提示：
1. 部分包含數(shù)學公式或PPT動畫的文件，查看預覽時可能會顯示錯亂或異常，文件下載后無此問題，請放心下載。
2. 本文檔由用戶上傳，版權歸屬用戶，天天文庫負責整理代發(fā)布。如果您對本文檔版權有爭議請及時聯(lián)系客服。
3. 下載前請仔細閱讀文檔內(nèi)容，確認文檔內(nèi)容符合您的需求后進行下載，若出現(xiàn)內(nèi)容與標題不符可向本站投訴處理。
4. 下載文檔時可能由于網(wǎng)絡波動等原因無法下載或下載錯誤，付費完成后未能成功下載的用戶請聯(lián)系客服處理。