NeDDF: 距離場と密度場を互恵的に制約する３次元形状表現

• Item type: SIG Technical Reports(1)
• 公開日: 2022-01-20
• タイトル: NeDDF: 距離場と密度場を互恵的に制約する３次元形状表現
• 言語: jpn
• キーワード / 主題Scheme: Other
• キーワード / 主題: 画像処理・機械学習
• 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: 筑波大学
• 著者所属: 早稲田大学
• 著者所属: 産業技術総合研究所
• 著者所属: 広島大学
• 著者所属: 筑波大学
• 著者所属: 筑波大学
• 著者名: 上田, 樹; 福原, 吉博; 片岡, 裕雄; 相澤, 宏旭; 宍戸, 英彦; 北原, 格
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: This paper proposes a novel 3D representation method, Neural Distance-Density Field (NeDDF), to constrain the distance and density fields reciprocally. Recently, several methods have been proposed to construct distance and density fields using Implicit Neural Representation (INR) for 3D reconstruction problems. However, methods using density fields such as NeRF do not provide density gradient in most free regions and thus have poor tracking performance for camera pose estimation. Also, applications of distance field methods such as NeuS are limited to object shapes with boundary surfaces. We propose a distance field representation that extends distances to non-boundary shapes and can be explicitly transformed into a density field.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: This paper proposes a novel 3D representation method, Neural Distance-Density Field (NeDDF), to constrain the distance and density fields reciprocally. Recently, several methods have been proposed to construct distance and density fields using Implicit Neural Representation (INR) for 3D reconstruction problems. However, methods using density fields such as NeRF do not provide density gradient in most free regions and thus have poor tracking performance for camera pose estimation. Also, applications of distance field methods such as NeuS are limited to object shapes with boundary surfaces. We propose a distance field representation that extends distances to non-boundary shapes and can be explicitly transformed into a density field.
• 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 2022-CVIM-228, 号 19, p. 1-6, 発行日 2022-01-20
• 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