返回
cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis

cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis

ID：40713785

大小：3.45 MB

页数：10页

时间：2019-08-06

cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis_第1页
cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis_第2页
cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis_第3页
cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis_第4页
cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis_第5页
资源描述：

《cvpr18-Two-Stream Convolutional Networks for Dynamic Texture Synthesis》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、Two-StreamConvolutionalNetworksforDynamicTextureSynthesisMatthewTesfaldetMarcusA.BrubakerKonstantinosG.DerpanisDepartmentofElectricalEngineeringandComputerScienceDepartmentofComputerScienceYorkUniversity,TorontoRyersonUniversity,Toronto{mtesfald,mab}@eecs.yorku.cakosta@scs.ryerson.caAbstractDynam

2、icTextureSynthesisDynamicsStyleTransferappearancetargetWeintroduceatwo-streammodelfordynamictextureappearance&dynamicstargetoutputsynthesis.Ourmodelisbasedonpre-trainedconvolutionaloutputnetworks(ConvNets)thattargettwoindependenttasks:(i)objectrecognition,and(ii)opticalflowprediction.Givenaninputd

3、ynamictexture,statisticsoffilterresponsesfromtheobjectrecognitionConvNetencapsulatetheper-frameappearanceoftheinputtexture,whilestatisticsoffilterre-sponsesfromtheopticalflowConvNetmodelitsdynamics.Togenerateanoveltexture,arandomlyinitializedinputse-dynamicstargetquenceisoptimizedtomatchthefeaturest

4、atisticsfromeachFigure1:Dynamictexturesynthesis.(left)Givenaninputstreamofanexampletexture.Inspiredbyrecentworkondynamictextureasthetarget,ourtwo-streammodelisableimagestyletransferandenabledbythetwo-streammodel,tosynthesizeanoveldynamictexturethatpreservesthetar-wealsoapplythesynthesisapproachto

5、combinethetextureget’sappearanceanddynamicscharacteristics.(right)Ourappearancefromonetexturewiththedynamicsofanothertwo-streamapproachenablessynthesisthatcombinesthetogenerateentirelynoveldynamictextures.Weshowthattextureappearancefromonetargetwiththedynamicsfromourapproachgeneratesnovel,highqua

6、litysamplesthatanother,resultinginacompositionofthetwo.matchboththeframewiseappearanceandtemporalevo-lutionofinputtexture.Finally,wequantitativelyevaluateexampletextureinputs,generatesanoveldynamictextureourtexturesynthesisapproachwithathoroughuserstudy.instance.Italsoenablesanovelformofstyletran

7、sferwherethetargetappearanceanddynamicscanbetakenfromdif-ferentsourcesasshowninFig.1.1.IntroductionOurmodelisconstructedfromtwoconvolutionalnet-works(ConvNets),anappearancestreamandadynamicsManycommontemporalvisualpatt

当前文档最多预览五页，下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页，下载文档查看全文
温馨提示：
1. 部分包含数学公式或PPT动画的文件，查看预览时可能会显示错乱或异常，文件下载后无此问题，请放心下载。
2. 本文档由用户上传，版权归属用户，天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容，确认文档内容符合您的需求后进行下载，若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误，付费完成后未能成功下载的用户请联系客服处理。