Unsupervised 3D Shape Completion through GAN Inversion

Most 3D shape completion approaches rely heavily on partial-complete shape pairs and learn in a fully supervised manner. Despite their impressive performances on in-domain data, when generalizing to partial shapes in other forms or real-world partial scans, they often obtain unsatisfactory results d...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Junzhe, Chen, Xinyi, Cai, Zhongang, Pan, Liang, Zhao, Haiyu, Yi, Shuai, Yeo, Chai Kiat, Dai, Bo, Loy, Chen Change
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Zhang, Junzhe
Chen, Xinyi
Cai, Zhongang
Pan, Liang
Zhao, Haiyu
Yi, Shuai
Yeo, Chai Kiat
Dai, Bo
Loy, Chen Change
description Most 3D shape completion approaches rely heavily on partial-complete shape pairs and learn in a fully supervised manner. Despite their impressive performances on in-domain data, when generalizing to partial shapes in other forms or real-world partial scans, they often obtain unsatisfactory results due to domain gaps. In contrast to previous fully supervised approaches, in this paper we present ShapeInversion, which introduces Generative Adversarial Network (GAN) inversion to shape completion for the first time. ShapeInversion uses a GAN pre-trained on complete shapes by searching for a latent code that gives a complete shape that best reconstructs the given partial input. In this way, ShapeInversion no longer needs paired training data, and is capable of incorporating the rich prior captured in a well-trained generative model. On the ShapeNet benchmark, the proposed ShapeInversion outperforms the SOTA unsupervised method, and is comparable with supervised methods that are learned using paired data. It also demonstrates remarkable generalization ability, giving robust results for real-world scans and partial inputs of various forms and incompleteness levels. Importantly, ShapeInversion naturally enables a series of additional abilities thanks to the involvement of a pre-trained GAN, such as producing multiple valid complete shapes for an ambiguous partial input, as well as shape manipulation and interpolation.
doi_str_mv 10.48550/arxiv.2104.13366
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2104_13366</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2104_13366</sourcerecordid><originalsourceid>FETCH-LOGICAL-a676-6bf68b8ce2174866eb9371cf0a0e70f99dd6ea3d1a44a440c6ddd4c18f8a57e03</originalsourceid><addsrcrecordid>eNotj8tuwjAURL1hgaAfwKr-gaQ2dq6dJUp5SahdFNbRTXxNIkESORC1f19e0kgjzeJoDmMzKWJtk0R8YPith3guhY6lUgBjBoemv3YUhronx9Un_6mwI5615-5El7pt-KUK7fVY8fXii2-bgUJ_W6ds5PHU09urJ2y_Wu6zTbT7Xm-zxS5CMBBB4cEWtqS5NNoCUJEqI0svUJARPk2dA0LlJGp9iyjBOadLab3FxJBQE_b-xD6O512ozxj-8rtA_hBQ_9gxQDA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Unsupervised 3D Shape Completion through GAN Inversion</title><source>arXiv.org</source><creator>Zhang, Junzhe ; Chen, Xinyi ; Cai, Zhongang ; Pan, Liang ; Zhao, Haiyu ; Yi, Shuai ; Yeo, Chai Kiat ; Dai, Bo ; Loy, Chen Change</creator><creatorcontrib>Zhang, Junzhe ; Chen, Xinyi ; Cai, Zhongang ; Pan, Liang ; Zhao, Haiyu ; Yi, Shuai ; Yeo, Chai Kiat ; Dai, Bo ; Loy, Chen Change</creatorcontrib><description>Most 3D shape completion approaches rely heavily on partial-complete shape pairs and learn in a fully supervised manner. Despite their impressive performances on in-domain data, when generalizing to partial shapes in other forms or real-world partial scans, they often obtain unsatisfactory results due to domain gaps. In contrast to previous fully supervised approaches, in this paper we present ShapeInversion, which introduces Generative Adversarial Network (GAN) inversion to shape completion for the first time. ShapeInversion uses a GAN pre-trained on complete shapes by searching for a latent code that gives a complete shape that best reconstructs the given partial input. In this way, ShapeInversion no longer needs paired training data, and is capable of incorporating the rich prior captured in a well-trained generative model. On the ShapeNet benchmark, the proposed ShapeInversion outperforms the SOTA unsupervised method, and is comparable with supervised methods that are learned using paired data. It also demonstrates remarkable generalization ability, giving robust results for real-world scans and partial inputs of various forms and incompleteness levels. Importantly, ShapeInversion naturally enables a series of additional abilities thanks to the involvement of a pre-trained GAN, such as producing multiple valid complete shapes for an ambiguous partial input, as well as shape manipulation and interpolation.</description><identifier>DOI: 10.48550/arxiv.2104.13366</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition</subject><creationdate>2021-04</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2104.13366$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2104.13366$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Zhang, Junzhe</creatorcontrib><creatorcontrib>Chen, Xinyi</creatorcontrib><creatorcontrib>Cai, Zhongang</creatorcontrib><creatorcontrib>Pan, Liang</creatorcontrib><creatorcontrib>Zhao, Haiyu</creatorcontrib><creatorcontrib>Yi, Shuai</creatorcontrib><creatorcontrib>Yeo, Chai Kiat</creatorcontrib><creatorcontrib>Dai, Bo</creatorcontrib><creatorcontrib>Loy, Chen Change</creatorcontrib><title>Unsupervised 3D Shape Completion through GAN Inversion</title><description>Most 3D shape completion approaches rely heavily on partial-complete shape pairs and learn in a fully supervised manner. Despite their impressive performances on in-domain data, when generalizing to partial shapes in other forms or real-world partial scans, they often obtain unsatisfactory results due to domain gaps. In contrast to previous fully supervised approaches, in this paper we present ShapeInversion, which introduces Generative Adversarial Network (GAN) inversion to shape completion for the first time. ShapeInversion uses a GAN pre-trained on complete shapes by searching for a latent code that gives a complete shape that best reconstructs the given partial input. In this way, ShapeInversion no longer needs paired training data, and is capable of incorporating the rich prior captured in a well-trained generative model. On the ShapeNet benchmark, the proposed ShapeInversion outperforms the SOTA unsupervised method, and is comparable with supervised methods that are learned using paired data. It also demonstrates remarkable generalization ability, giving robust results for real-world scans and partial inputs of various forms and incompleteness levels. Importantly, ShapeInversion naturally enables a series of additional abilities thanks to the involvement of a pre-trained GAN, such as producing multiple valid complete shapes for an ambiguous partial input, as well as shape manipulation and interpolation.</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8tuwjAURL1hgaAfwKr-gaQ2dq6dJUp5SahdFNbRTXxNIkESORC1f19e0kgjzeJoDmMzKWJtk0R8YPith3guhY6lUgBjBoemv3YUhronx9Un_6mwI5615-5El7pt-KUK7fVY8fXii2-bgUJ_W6ds5PHU09urJ2y_Wu6zTbT7Xm-zxS5CMBBB4cEWtqS5NNoCUJEqI0svUJARPk2dA0LlJGp9iyjBOadLab3FxJBQE_b-xD6O512ozxj-8rtA_hBQ_9gxQDA</recordid><startdate>20210427</startdate><enddate>20210427</enddate><creator>Zhang, Junzhe</creator><creator>Chen, Xinyi</creator><creator>Cai, Zhongang</creator><creator>Pan, Liang</creator><creator>Zhao, Haiyu</creator><creator>Yi, Shuai</creator><creator>Yeo, Chai Kiat</creator><creator>Dai, Bo</creator><creator>Loy, Chen Change</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20210427</creationdate><title>Unsupervised 3D Shape Completion through GAN Inversion</title><author>Zhang, Junzhe ; Chen, Xinyi ; Cai, Zhongang ; Pan, Liang ; Zhao, Haiyu ; Yi, Shuai ; Yeo, Chai Kiat ; Dai, Bo ; Loy, Chen Change</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a676-6bf68b8ce2174866eb9371cf0a0e70f99dd6ea3d1a44a440c6ddd4c18f8a57e03</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><toplevel>online_resources</toplevel><creatorcontrib>Zhang, Junzhe</creatorcontrib><creatorcontrib>Chen, Xinyi</creatorcontrib><creatorcontrib>Cai, Zhongang</creatorcontrib><creatorcontrib>Pan, Liang</creatorcontrib><creatorcontrib>Zhao, Haiyu</creatorcontrib><creatorcontrib>Yi, Shuai</creatorcontrib><creatorcontrib>Yeo, Chai Kiat</creatorcontrib><creatorcontrib>Dai, Bo</creatorcontrib><creatorcontrib>Loy, Chen Change</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Zhang, Junzhe</au><au>Chen, Xinyi</au><au>Cai, Zhongang</au><au>Pan, Liang</au><au>Zhao, Haiyu</au><au>Yi, Shuai</au><au>Yeo, Chai Kiat</au><au>Dai, Bo</au><au>Loy, Chen Change</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Unsupervised 3D Shape Completion through GAN Inversion</atitle><date>2021-04-27</date><risdate>2021</risdate><abstract>Most 3D shape completion approaches rely heavily on partial-complete shape pairs and learn in a fully supervised manner. Despite their impressive performances on in-domain data, when generalizing to partial shapes in other forms or real-world partial scans, they often obtain unsatisfactory results due to domain gaps. In contrast to previous fully supervised approaches, in this paper we present ShapeInversion, which introduces Generative Adversarial Network (GAN) inversion to shape completion for the first time. ShapeInversion uses a GAN pre-trained on complete shapes by searching for a latent code that gives a complete shape that best reconstructs the given partial input. In this way, ShapeInversion no longer needs paired training data, and is capable of incorporating the rich prior captured in a well-trained generative model. On the ShapeNet benchmark, the proposed ShapeInversion outperforms the SOTA unsupervised method, and is comparable with supervised methods that are learned using paired data. It also demonstrates remarkable generalization ability, giving robust results for real-world scans and partial inputs of various forms and incompleteness levels. Importantly, ShapeInversion naturally enables a series of additional abilities thanks to the involvement of a pre-trained GAN, such as producing multiple valid complete shapes for an ambiguous partial input, as well as shape manipulation and interpolation.</abstract><doi>10.48550/arxiv.2104.13366</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2104.13366
ispartof
issn
language eng
recordid cdi_arxiv_primary_2104_13366
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
title Unsupervised 3D Shape Completion through GAN Inversion
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-23T20%3A36%3A03IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Unsupervised%203D%20Shape%20Completion%20through%20GAN%20Inversion&rft.au=Zhang,%20Junzhe&rft.date=2021-04-27&rft_id=info:doi/10.48550/arxiv.2104.13366&rft_dat=%3Carxiv_GOX%3E2104_13366%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true