{"created":"2025-01-19T01:29:16.843383+00:00","updated":"2025-01-19T11:23:12.984633+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00229863","sets":["6504:11436:11440"]},"path":["11440"],"owner":"44499","recid":"229863","title":["複数解像度で画像を生成可能な拡散確率モデル"],"pubdate":{"attribute_name":"公開日","attribute_value":"2023-02-16"},"_buckets":{"deposit":"66bdece2-c1ec-4857-907f-69975303d180"},"_deposit":{"id":"229863","pid":{"type":"depid","value":"229863","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"複数解像度で画像を生成可能な拡散確率モデル","author_link":["618342","618345","618343","618344"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"複数解像度で画像を生成可能な拡散確率モデル"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"人工知能と認知科学","subitem_subject_scheme":"Other"}]},"item_type_id":"22","publish_date":"2023-02-16","item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_22_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"早大"},{"subitem_text_value":"早大"},{"subitem_text_value":"東大"},{"subitem_text_value":"早大"}]},"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/229863/files/IPSJ-Z85-2P-08.pdf","label":"IPSJ-Z85-2P-08.pdf"},"date":[{"dateType":"Available","dateValue":"2023-11-17"}],"format":"application/pdf","filename":"IPSJ-Z85-2P-08.pdf","filesize":[{"value":"808.1 kB"}],"mimetype":"application/pdf","accessrole":"open_date","version_id":"3edec251-a2da-40c0-9be6-b05676544aee","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2023 by the Information Processing Society of Japan"}]},"item_22_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"荒川, 深映"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"綱島, 秀樹"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"堀田, 大地"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"森島, 繁生"}],"nameIdentifiers":[{}]}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_5794","resourcetype":"conference paper"}]},"item_22_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN00349328","subitem_source_identifier_type":"NCID"}]},"item_22_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"拡散モデルは画像生成において最先端の性能を示している．一方で，固定した解像度で訓練された拡散モデルは推論時に解像度を変更すると，画像全体の構造やテクスチャが崩壊し，低品質となる問題を持つ．そこで本研究では，複数解像度での高品質な画像生成を可能とする学習法を提案する．学習時に複数解像度の画像から切り抜いたパッチを用いることによって，画像全体の構造やディテールを学習し，様々な解像度での生成を可能にする．この学習法によって，推論時に学習時よりも高解像度の画像を生成可能であることに加えて，学習に低解像度のパッチを用いることが計算コストの削減に繋がることを明らかにした．","subitem_description_type":"Other"}]},"item_22_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"126","bibliographic_titles":[{"bibliographic_title":"第85回全国大会講演論文集"}],"bibliographicPageStart":"125","bibliographicIssueDates":{"bibliographicIssueDate":"2023-02-16","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"1","bibliographicVolumeNumber":"2023"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":229863,"links":{}}