报告预览

4-4 敏捷型风格化人像和虚拟形象创建.pdf

编号：102505

PDF 50页 30.08MB 下载积分：VIP专享

下载报告请您先登录！

4-4 敏捷型风格化人像和虚拟形象创建.pdf

1、Agile Stylized Portrait and Avatar Creation敏捷型风格化人像和虚拟形象创建罗琳捷数字人技术北美研究主管The Rising Popularity of Virtual Identities2Viola app/Snapchat cartoon filter Stylized PortraitsMemoji/ReadyPlayerMe/Bitmoji Stylized AvatarsVirtual Identities in ByteDance Products3抖音仔仔Manga StylePixar StyleTikTok Avatars4How

2、to automatically createlook-alike&appealing stylized virtual identities for usersExisting Approach Requires Laborious Annotations5large amount of paired stylized examplesFace shapeNose shapeHair lenghHair colorGlassesPearNubianLongYellowYesTriangleRomanShortBrownNoOvalSnubLongBlackYeslarge amount of

3、 annotationsKey ChallengesHow to reduce or eliminate the manual effort and increase agility?6Lower costFaster ShippingHigher QualityStylized Portraits7AgileGAN:Stylizing Portraits via Inversion-Consistent Transfer Learning SIGGRAPH 2021AgileAvatar:Stylized 3D Avatar Creation via Cascaded Domain Brid

4、ging SIGGRAPH Asia 2022Stylized AvatarsStylized Portraits8AgileGAN:Stylizing Portraits via Inversion-Consistent Transfer Learning SIGGRAPH 2021AgileAvatar:Stylized 3D Avatar Creation via Cascaded Domain Bridging SIGGRAPH Asia 2022Stylized AvatarsPrevious Work Manual Creation9Top quality,but very exp

5、ensivePrevious Work Style Transfer10InputGatys et al.2016Style exemplarLack semantic correspondencesPrevious Work Unsupervised Image Translation11InputCycleGAN Zhu et al.2017UGATIT Kim et al.2020UNIT Liu et al.2017Sub-optimal quality due to lack of style exemplarsPrevious Work ToonifyTransfer learni

6、ng from StyleGAN by very few(100)style exemplarsInputToonifyInputToonifyStyle exemplarsPrevious Work ToonifyTransfer learning from StyleGAN by very few(100)style exemplars13InputInputOur AgileGAN ResultsPortrait Stylization by Transfer Learning from StyleGAN14Z space(1 x 512)W space(1 x 512)MLPStyle

7、GAN2 generatorPortrait Stylization by Transfer Learning from StyleGAN15Z space(1 x 512)W space(1 x 512)MLPStyleGAN2 generatorDiscriminatorPortrait Stylization by Transfer Learning from StyleGAN16Z space(1 x 512)W space(1 x 512)MLPTransfer-learned generatorDiscriminator 2021 SIGGRAPH.ALL RIGHTS RESER

8、VED.Training II:Generator17Augmentation:Multi-Resolution Latent Spaces:Z+/W+18Z+space(18 x 512)W+space(18 x 512)MLPTransfer-learned generatorDiscriminatorInputZ/W spacesZ+/W+spacesThe Problem of StyleGAN Inversion19MLPStyleGAN2 generatorTransfer-learned generatorDiscriminator?Z+space(18 x 512)W+spac

9、e(18 x 512)The Problem of StyleGAN Inversion20MLPStyleGAN2 generatorTransfer-learned generatorDiscriminator?Z+space(18 x 512)W+space(18 x 512)W+Space Latent Code Distribution(tSNE)Training data distributionThe Problem of StyleGAN Inversion21MLPStyleGAN2 generatorTransfer-learned generatorDiscriminat

10、or?Z+space(18 x 512)W+space(18 x 512)W+Space Latent Code Distribution(tSNE)Toonify(optimization)In-domain Zhu et al.2020PSP Richardson et al.2020Training data distributionThe Problem of StyleGAN Inversion22MLPStyleGAN2 generatorTransfer-learned generatorDiscriminator?Z+space(18 x 512)W+space(18 x 51

11、2)W+Space Latent Code Distribution(tSNE)Toonify(optimization)Our methodIn-domain Zhu et al.2020PSP Richardson et al.2020Training data distribution23242526Hierarchical Variational Autoencoder(hVAE)27MLPZ+space(18 x 512)W+space(18 x 512)StyleGAN2 generatorHierarchical Variational Autoencoder(hVAE)28ML

12、PZ+space(18 x 512)W+space(18 x 512)StyleGAN2 generator 2021 SIGGRAPH.ALL RIGHTS RESERVED.Stylization Results29MORE STYLES30More Styles 2021 SIGGRAPH.ALL RIGHTS RESERVED.Comparisons for State of the art31Limitations32Stylized Portraits33AgileGAN:Stylizing Portraits via Inversion-Consistent Transfer L

13、earning SIGGRAPH 2021AgileAvatar:Stylized 3D Avatar Creation via Cascaded Domain Bridging SIGGRAPH Asia 2022Stylized AvatarsPrevious work34Shi et al.Fast and Robust Face-to-Parameter Translation for Game Character Auto-Creation.AAAI 2020Previous work35Shi et al.Fast and Robust Face-to-Parameter Tran

14、slation for Game Character Auto-Creation.AAAI 2020Previous work36Shi et al.Fast and Robust Face-to-Parameter Translation for Game Character Auto-Creation.AAAI 2020Previous work37Shi et al.Fast and Robust Face-to-Parameter Translation for Game Character Auto-Creation.AAAI 2020Enabled end-to-end train

15、ing!Previous work38Shi et al.Fast and Robust Face-to-Parameter Translation for Game Character Auto-Creation.AAAI 2020Previous work39Shi et al.Fast and Robust Face-to-Parameter Translation for Game Character Auto-Creation.AAAI 2020Realistic Style Only!Motivation40AgileGANLossApproach41Improve AgileGA

16、Ns Stylization42Normalized Style ExemplarsPreserves identity features(e.g.glasses)and neutralize expressionsTricks to Optimize Discrete ParametersSoft(relaxed)parameter assignment is important for optimization to converge43Tricks to Optimize Discrete ParametersSoft(relaxed)parameter assignment is im

17、portant for optimization to convergeDirectly quantizing soft parameter assignments to hard values can be sub-optimal a search-based technique is needed to retain visual similarity.4445Ours F2P Shi et al.2019 ManualComparisonUser Study46Attribute Evaluation:the percentage of users think the specific

18、output attribute matches inputMatch Task:the percentage of users think the output avatar matches inputConclusion:Our method is superior to competing methods and comparable to manual resultsLimitations47Stylized Portraits48AgileGAN:Stylizing Portraits via Inversion-Consistent Transfer Learning SIGGRA

19、PH 2021AgileAvatar:Stylized 3D Avatar Creation via Cascaded Domain Bridging SIGGRAPH Asia 2022Stylized AvatarsStylized Portraits49AgileGAN:Stylizing Portraits via Inversion-Consistent Transfer Learning SIGGRAPH 2021AgileAvatar:Stylized 3D Avatar Creation via Cascaded Domain Bridging SIGGRAPH Asia 20

20、22Stylized AvatarsWe can develop portrait stylization models from very few(100)unpaired style exemplars,and speed up product deployment from months to weeksWe can develop auto avatar creation models without laborious annotations,and quickly adapt to different avatar styles and system changesThank you!50

友情提示

1、下载报告失败解决办法
2、PDF文件下载后，可能会被浏览器默认打开，此种情况可以点击浏览器菜单，保存网页到桌面，就可以正常下载了。
3、本站不支持迅雷下载，请使用电脑自带的IE浏览器，或者360浏览器、谷歌浏览器下载即可。
4、本站报告下载后的文档和图纸-无水印,预览文档经过压缩，下载后原文更清晰。

本文（4-4 敏捷型风格化人像和虚拟形象创建.pdf）为本站（云闲）主动上传，三个皮匠报告文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若此文所含内容侵犯了您的版权或隐私，请立即通知三个皮匠报告文库（点击联系客服），我们立即给予删除！

温馨提示：如果因为网速或其他原因下载失败请重新下载，重复下载不扣分。

上海品茶

4-4 敏捷型风格化人像和虚拟形象创建.pdf

4-4 敏捷型风格化人像和虚拟形象创建.pdf