GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields

We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene cons...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sekikawa, Yusuke, Hsu, Chingwei, Ikehata, Satoshi, Kawakami, Rei, Sato, Ikuro
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 Sekikawa, Yusuke
Hsu, Chingwei
Ikehata, Satoshi
Kawakami, Rei
Sato, Ikuro
description We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene consisting of many objects. However, we observe that such a MoE NeRF model often produces low-quality representations in the vicinity of experts' boundaries when applied to the task of novel view synthesis of an unseen object from one/few-shot input. We find that this deterioration is primarily caused by the foresight expert selection mechanism, which may leave an unnatural discontinuity in the object shape near the experts' boundaries. Gumbel-NeRF adopts a hindsight expert selection mechanism, which guarantees continuity in the density field even near the experts' boundaries. Experiments using the SRN cars dataset demonstrate the superiority of Gumbel-NeRF over the baselines in terms of various image quality metrics.
doi_str_mv 10.48550/arxiv.2410.20306
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2410_20306</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2410_20306</sourcerecordid><originalsourceid>FETCH-arxiv_primary_2410_203063</originalsourceid><addsrcrecordid>eNpjYJA0NNAzsTA1NdBPLKrILNMzMgEKGBkYG5hxMoS6h_o6ufro-rkGuVkpBKUWFKUWp-aVZOalK4TmFaem5in4J2WlJpcUKyQWKwQkFpXoOufnFuQXZ5Zk5ucl5ij4pZYWAamgxJTMxLzkVAW3zNSclGIeBta0xJziVF4ozc0g7-Ya4uyhC3ZAfEFRZm5iUWU8yCHxYIcYE1YBAH_APQk</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields</title><source>arXiv.org</source><creator>Sekikawa, Yusuke ; Hsu, Chingwei ; Ikehata, Satoshi ; Kawakami, Rei ; Sato, Ikuro</creator><creatorcontrib>Sekikawa, Yusuke ; Hsu, Chingwei ; Ikehata, Satoshi ; Kawakami, Rei ; Sato, Ikuro</creatorcontrib><description>We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene consisting of many objects. However, we observe that such a MoE NeRF model often produces low-quality representations in the vicinity of experts' boundaries when applied to the task of novel view synthesis of an unseen object from one/few-shot input. We find that this deterioration is primarily caused by the foresight expert selection mechanism, which may leave an unnatural discontinuity in the object shape near the experts' boundaries. Gumbel-NeRF adopts a hindsight expert selection mechanism, which guarantees continuity in the density field even near the experts' boundaries. Experiments using the SRN cars dataset demonstrate the superiority of Gumbel-NeRF over the baselines in terms of various image quality metrics.</description><identifier>DOI: 10.48550/arxiv.2410.20306</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition</subject><creationdate>2024-10</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,781,886</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2410.20306$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2410.20306$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Sekikawa, Yusuke</creatorcontrib><creatorcontrib>Hsu, Chingwei</creatorcontrib><creatorcontrib>Ikehata, Satoshi</creatorcontrib><creatorcontrib>Kawakami, Rei</creatorcontrib><creatorcontrib>Sato, Ikuro</creatorcontrib><title>GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields</title><description>We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene consisting of many objects. However, we observe that such a MoE NeRF model often produces low-quality representations in the vicinity of experts' boundaries when applied to the task of novel view synthesis of an unseen object from one/few-shot input. We find that this deterioration is primarily caused by the foresight expert selection mechanism, which may leave an unnatural discontinuity in the object shape near the experts' boundaries. Gumbel-NeRF adopts a hindsight expert selection mechanism, which guarantees continuity in the density field even near the experts' boundaries. Experiments using the SRN cars dataset demonstrate the superiority of Gumbel-NeRF over the baselines in terms of various image quality metrics.</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNpjYJA0NNAzsTA1NdBPLKrILNMzMgEKGBkYG5hxMoS6h_o6ufro-rkGuVkpBKUWFKUWp-aVZOalK4TmFaem5in4J2WlJpcUKyQWKwQkFpXoOufnFuQXZ5Zk5ucl5ij4pZYWAamgxJTMxLzkVAW3zNSclGIeBta0xJziVF4ozc0g7-Ya4uyhC3ZAfEFRZm5iUWU8yCHxYIcYE1YBAH_APQk</recordid><startdate>20241026</startdate><enddate>20241026</enddate><creator>Sekikawa, Yusuke</creator><creator>Hsu, Chingwei</creator><creator>Ikehata, Satoshi</creator><creator>Kawakami, Rei</creator><creator>Sato, Ikuro</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20241026</creationdate><title>GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields</title><author>Sekikawa, Yusuke ; Hsu, Chingwei ; Ikehata, Satoshi ; Kawakami, Rei ; Sato, Ikuro</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-arxiv_primary_2410_203063</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><toplevel>online_resources</toplevel><creatorcontrib>Sekikawa, Yusuke</creatorcontrib><creatorcontrib>Hsu, Chingwei</creatorcontrib><creatorcontrib>Ikehata, Satoshi</creatorcontrib><creatorcontrib>Kawakami, Rei</creatorcontrib><creatorcontrib>Sato, Ikuro</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Sekikawa, Yusuke</au><au>Hsu, Chingwei</au><au>Ikehata, Satoshi</au><au>Kawakami, Rei</au><au>Sato, Ikuro</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields</atitle><date>2024-10-26</date><risdate>2024</risdate><abstract>We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene consisting of many objects. However, we observe that such a MoE NeRF model often produces low-quality representations in the vicinity of experts' boundaries when applied to the task of novel view synthesis of an unseen object from one/few-shot input. We find that this deterioration is primarily caused by the foresight expert selection mechanism, which may leave an unnatural discontinuity in the object shape near the experts' boundaries. Gumbel-NeRF adopts a hindsight expert selection mechanism, which guarantees continuity in the density field even near the experts' boundaries. Experiments using the SRN cars dataset demonstrate the superiority of Gumbel-NeRF over the baselines in terms of various image quality metrics.</abstract><doi>10.48550/arxiv.2410.20306</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2410.20306
ispartof
issn
language eng
recordid cdi_arxiv_primary_2410_20306
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
title GUMBEL-NERF: Representing Unseen Objects as Part-Compositional Neural Radiance Fields
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-18T14%3A54%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=GUMBEL-NERF:%20Representing%20Unseen%20Objects%20as%20Part-Compositional%20Neural%20Radiance%20Fields&rft.au=Sekikawa,%20Yusuke&rft.date=2024-10-26&rft_id=info:doi/10.48550/arxiv.2410.20306&rft_dat=%3Carxiv_GOX%3E2410_20306%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