{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00225160","sets":["1164:2240:11176:11177"]},"path":["11177"],"owner":"44499","recid":"225160","title":["マルチGPUによる2次元画像超解像モデルの性能評価"],"pubdate":{"attribute_name":"公開日","attribute_value":"2023-03-09"},"_buckets":{"deposit":"967c119a-97af-4aa2-9796-26d506ad7d75"},"_deposit":{"id":"225160","pid":{"type":"depid","value":"225160","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"マルチGPUによる2次元画像超解像モデルの性能評価","author_link":["595297","595298","595299","595300"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"マルチGPUによる2次元画像超解像モデルの性能評価"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"GPU","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2023-03-09","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"神戸大学工学部"},{"subitem_text_value":"神戸大学大学院システム情報学研究科"},{"subitem_text_value":"岡山大学大学院生命科学研究科"},{"subitem_text_value":"神戸大学大学院システム情報学研究科"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Faculty of Engineering, Kobe University","subitem_text_language":"en"},{"subitem_text_value":"Graduate school of System Informatics, Kobe University","subitem_text_language":"en"},{"subitem_text_value":"Graduate school of Environmental and Life Science, Okayama University","subitem_text_language":"en"},{"subitem_text_value":"Graduate school of System Informatics, Kobe University","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/225160/files/IPSJ-HPC23188016.pdf","label":"IPSJ-HPC23188016.pdf"},"date":[{"dateType":"Available","dateValue":"2025-03-09"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-HPC23188016.pdf","filesize":[{"value":"4.3 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"14"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"94f9c010-d936-42e6-9585-fbfb6f102d61","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2023 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"福井, 匠"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"鈴木, 綾介"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"石原, 卓"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"横川, 三津夫"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10463942","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8841","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"データの再構築はデータサイエンス分野において重要な課題である．画像データの再構築の方法の一つである超解像は，低解像データから高解像データを構築する手法であり，近年では，深層学習を用いた超解像技術が流体力学分野に応用されはじめた．しかし，これまで 2 次元流れ場画像の超解像は盛んに研究されているが，計算コストが大きいため 3 次元画像への超解像の適用例は少ない．本研究では，既存の複数の超解像モデルについて誤差と実行時間についての性能をスーパーコンピュータ「不老 Type2」を用いて評価した．また，最も結果の良かったモデル Very deep residual channel attention networks (RCAN) を用いて，マルチ GPU 環境下で分散学習を行い，並列度を変えて，誤差と実行時間についての性能を評価した．誤差については，GPU 数を増やすと局所解から大きく離れてしまうことがあるが，学習率を調整することで抑えられることが分かった．実行時間については，GPU 数が 32 の時に GPU 数が 1 の時に比べて約 22 倍の高速化が得られた．さらに，2 次元超解像モデル SRResNet を拡張した 3 次元超解像モデル 3D-SRResNet を作成し，3 次元渦度画像の学習データ 80 枚に対し分散学習を行い，並列処理により実行時間が大幅に短縮できること，一定の超解像ができることを明らかにした．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告ハイパフォーマンスコンピューティング（HPC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2023-03-09","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"16","bibliographicVolumeNumber":"2023-HPC-188"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":225160,"updated":"2025-01-19T12:54:05.013042+00:00","links":{},"created":"2025-01-19T01:24:40.285262+00:00"}