Generalizable Novel-view Synthesis of Full-body Human from Sparse Input

• Item type: SIG Technical Reports(1)
• 公開日: 2023-11-09
• タイトル: Generalizable Novel-view Synthesis of Full-body Human from Sparse Input
• タイトル / 言語: en
• タイトル: Generalizable Novel-view Synthesis of Full-body Human from Sparse Input
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Presently with University of Tsukuba
• 著者所属: Presently with University of Tsukuba
• 著者所属: Presently with University of Tsukuba
• 著者所属(英): en
• 著者所属(英): Presently with University of Tsukuba
• 著者所属(英): en
• 著者所属(英): Presently with University of Tsukuba
• 著者所属(英): en
• 著者所属(英): Presently with University of Tsukuba
• 著者名: Zhaorong, Wang; Yoshihiro, Kanamori; Yuki, Endo
• 著者名(英): Zhaorong, Wang; Yoshihiro, Kanamori; Yuki, Endo
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Neural Radiance Fields (NeRF) have significantly advanced the field of novel view synthesis, and its application to full-body humans (or human NeRF) has been deemed as promising to enable telepresence. Generalizable human NeRF can avoid lengthy re-training for each human target but, if only sparse views are given, suffers from blurry outputs with artifacts due to insufficient visible samples. To better handle such a sparse view setting, we enhance the quality of appearance particularly in the regions completely occluded in any input views. We first condense sampling rays by omitting empty spaces via a parametric body fitting, leading improved appearance. We then specify the completely occluded regions and inpaint them to remove artifacts. Our method demonstrates improvements in quantitative evaluations compared to the baseline method. Qualitative results also exhibit higher fidelity, fewer artifacts, and a more natural clothing appearance.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Neural Radiance Fields (NeRF) have significantly advanced the field of novel view synthesis, and its application to full-body humans (or human NeRF) has been deemed as promising to enable telepresence. Generalizable human NeRF can avoid lengthy re-training for each human target but, if only sparse views are given, suffers from blurry outputs with artifacts due to insufficient visible samples. To better handle such a sparse view setting, we enhance the quality of appearance particularly in the regions completely occluded in any input views. We first condense sampling rays by omitting empty spaces via a parametric body fitting, leading improved appearance. We then specify the completely occluded regions and inpaint them to remove artifacts. Our method demonstrates improvements in quantitative evaluations compared to the baseline method. Qualitative results also exhibit higher fidelity, fewer artifacts, and a more natural clothing appearance.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2023-CVIM-235, 号 16, p. 1-6, 発行日 2023-11-09
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8701
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会