{"updated":"2025-01-19T20:46:28.598689+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00202766","sets":["1164:3696:10077:10078"]},"path":["10078"],"owner":"44499","recid":"202766","title":["敵対的生成ネットワーク（GAN）を用いた似顔絵生成手法の検討"],"pubdate":{"attribute_name":"公開日","attribute_value":"2020-01-16"},"_buckets":{"deposit":"178102a4-07c8-4bed-88a3-0d56da4b1af5"},"_deposit":{"id":"202766","pid":{"type":"depid","value":"202766","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"敵対的生成ネットワーク（GAN）を用いた似顔絵生成手法の検討","author_link":["498365","498367","498366","498368"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"敵対的生成ネットワーク（GAN）を用いた似顔絵生成手法の検討"},{"subitem_title":"A generation method of cartoon portrait using Generative Adversarial Networks","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"DCC","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2020-01-16","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"神奈川工科大学大学院情報工学専攻"},{"subitem_text_value":"神奈川工科大学情報メディア学科"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Kanagawa Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"Kanagawa Institute of Technology","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/202766/files/IPSJ-GN20109029.pdf","label":"IPSJ-GN20109029.pdf"},"date":[{"dateType":"Available","dateValue":"2022-01-16"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-GN20109029.pdf","filesize":[{"value":"1.8 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":"29"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"1d21da6f-2210-4157-a7fa-b92465c74952","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2020 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"中島, 悠輔"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"坂内, 祐一"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yusuke, Nakashima","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yuichi, Bannai","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AA1155524X","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":"2188-8744","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"似顔絵は人物の外見・特徴をとらえて，デフォルメして描いた人物画である．現在，顔画像を似顔絵に変換する研究では，非教師学習を用いた変換手法や，それぞれパーツ毎に変換を行う手法の研究が行われている．しかし，プロのイラストレーターの個性を反映させるような研究は発表されていない．そこで，本研究ではプロのイラストレーターの個性を反映する為に，どのネットワークを用いれば良いか検討を行う．具体的には，pix2pix，CycleGAN ペア，CycleGAN 非ペア，Cyclepix の 4 つのネットワークを検討していく．違いとしては主に損失関数となっている．pix2pix は訓練データと生成データの誤差を取るが，CycleGAN では入力データと生成データを更に変換させた再変換データの誤差を取ることが主な違いである．Cyclepix は両方の誤差を取っている．また，pix2pix，CycleGAN ペアは Discriminator の入力が入力データと生成データペアと条件付けされており，CycleGAN 非ペアと Cyclepix では Discriminator の入力が生成データのみとなっていることも違いである．実験結果として，CycleGAN 非ペアと Cyclepix の評価が高いことが分かった．このことから Discriminator の入力が生成データのみ，Cycle Consistency Loss を利用することで精度が高い似顔絵を生成することに有用である．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"6","bibliographic_titles":[{"bibliographic_title":"研究報告グループウェアとネットワークサービス（GN）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2020-01-16","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"29","bibliographicVolumeNumber":"2020-GN-109"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"created":"2025-01-19T01:05:18.696103+00:00","id":202766,"links":{}}