返回
Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks

Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks

ID：40726094

大小：5.49 MB

页数：9页

时间：2019-08-06

Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks_第1页
Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks_第2页
Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks_第3页
Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks_第4页
Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks_第5页
资源描述：

《Shape Inpainting Using 3D Generative Adversarial Network and Recurrent Convolutional Networks》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、ShapeInpaintingusing3DGenerativeAdversarialNetworkandRecurrentConvolutionalNetworksWeiyueWang1QianguiHuang1SuyaYou2ChaoYang1UlrichNeumann11UniversityofSouthernCalifornia2USArmyResearchLaboratoryLosAngeles,CaliforniaPlayaVista,California{weiyuewa,qianguih,chaoy,uneumann}@usc.edusu

2、ya.you.civ@mail.milAbstractgenerationandimageinpainting.Generatingandinpaint-ing3DmodelsisanewandmorechallengingproblemdueRecentadvancesinconvolutionalneuralnetworkshavetoitshigherdimensionality.Theavailabilityoflarge3Dshownpromisingresultsin3Dshapecompletion.ButduetoCADdatasets[

3、5,27]andCNNsforvoxel(spatialoccu-GPUmemorylimitations,thesemethodscanonlyproducepancy)models[26,21,8]enabledprogressinlearning3Dlow-resolutionoutputs.Toinpaint3Dmodelswithseman-representation,shapegenerationandcompletion.Despiteticplausibilityandcontextualdetails,weintroduceahy-t

4、heirencouragingresults,artifactsstillpersistsintheirgen-bridframeworkthatcombinesa3DEncoder-DecoderGen-eratedshapes.Moreover,theirmethodsareallbasedon3DerativeAdversarialNetwork(3D-ED-GAN)andaLong-CNN,whichimpedestheirabilitytohandlehigherresolu-termRecurrentConvolutionalNetwork(

5、LRCN).The3D-tiondataduetolimitedGPUmemory.ED-GANisa3DconvolutionalneuralnetworktrainedwithInthispaper,anewsystemfor3Dobjectinpaintingagenerativeadversarialparadigmtofillmissing3Ddataisintroducedtoovercometheaforementionedlimitations.inlow-resolution.LRCNadoptsarecurrentneuralnet-G

6、ivena3Dobjectwithholes,weaimto(1)fillthemissingworkarchitecturetominimizeGPUmemoryusageandin-ordamagedportionsandreconstructacomplete3Dstruc-corporatesanEncoder-DecoderpairintoaLongShort-ture,and(2)furtherpredicthigh-resolutionshapeswithtermMemoryNetwork.Byhandlingthe3Dmodelasase-

7、fine-graineddetails.Weproposeahybridnetworkstructurequenceof2Dslices,LRCNtransformsacoarse3Dshapebasedon3DCNNthatleveragesthegeneralizationpowerintoamorecompleteandhigherresolutionvolume.WhileofaGenerativeAdversarialmodelandthememoryeffi-3D-ED-GANcapturesglobalcontextualstructureof

8、the3DciencyofRecurrentNeuralNetwork(RNN)

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。