{"links":{},"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00225950","sets":["1164:4619:11188:11274"]},"path":["11274"],"owner":"44499","recid":"225950","title":["動画フレーム補間に特化した教師オプティカルフローの作成"],"pubdate":{"attribute_name":"公開日","attribute_value":"2023-05-11"},"_buckets":{"deposit":"bd750462-777c-470b-be2f-c9183b4980b6"},"_deposit":{"id":"225950","pid":{"type":"depid","value":"225950","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"動画フレーム補間に特化した教師オプティカルフローの作成","author_link":["599091","599088","599092","599089","599090","599093"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"動画フレーム補間に特化した教師オプティカルフローの作成"},{"subitem_title":"Creation of Teacher Optical Flow specialized for Video Frame Interpolation","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"セッション1","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2023-05-11","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"豊田工業大学"},{"subitem_text_value":"豊田工業大学"},{"subitem_text_value":"豊田工業大学"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Toyota Technological Institute","subitem_text_language":"en"},{"subitem_text_value":"Toyota Technological Institute","subitem_text_language":"en"},{"subitem_text_value":"Toyota Technological Institute","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/225950/files/IPSJ-CVIM23234006.pdf","label":"IPSJ-CVIM23234006.pdf"},"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-CVIM23234006.pdf","filesize":[{"value":"2.6 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"0","billingrole":"20"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_login","version_id":"876e76c1-ad3d-4b9a-a749-62fec3d8253d","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2023 by the Institute of Electronics, Information and Communication Engineers This SIG report is only available to those in membership of the SIG."}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"宮田, 陸"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"近藤, 佑樹"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"浮田, 宗伯"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Riku, Miyata","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yuki, Kondo","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Norimichi, Ukita","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AA11131797","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":"2188-8701","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"動画フレーム補間とは，連続する 2 つの動画フレームから中間のフレームを生成する技術である．本研究では，知識蒸留という学習戦略を利用し，高品質な動画フレーム補間を実現することを目指す．知識蒸留は，教師モデルから生徒モデルへの知識の伝達を促す手法である．この研究で扱う従来手法は RIFE [1] であり，動画フレーム補間タスクで高い性能を示している．RIFE とは異なり，本研究ではより大きな教師モデルを使用することを提案する．具体的には，オプティカルフローの推定において，SOTA（State-Of-The-Art）性能を持つ FlowFormer [2] モデルを教師モデルとして採用する．そうして得られたオプティカルフローを複数回 U-Net [3] に通すことによってさらに，精度の良いオプティカルフローを作成すること期待する．実験ではまず，教師モデルを作成し，オプティカルフローを推定する．その後，教師モデルの推定オプティカルフローを疑似 GT（Ground Truth）として利用し，生徒モデルを学習させる．実験の結果，提案手法と RIFE を精度で比較したが，予想に反して提案手法の精度は RIFE よりも低かった．これは，教師モデルを単純に巨大化するだけでは，必ずしも生徒モデルの精度向上につながらないことを示している．この結果から，動画フレーム補間タスクにおいて，さらなる工夫が必要であることが明らかとなった．","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Video frame interpolation is a technique for generating intermediate frames from two consecutive video frames. In this study, we aim to achieve high-quality video frame interpolation using a learning strategy called knowledge distillation. Knowledge distillation is a method to promote the transfer of knowledge from a teacher model to a student model. The conventional method used in this study is RIFE [1], which demonstrates high performance in video frame interpolation tasks. Unlike RIFE, this study proposes to use a larger teacher model. Specifically, we adopt the FlowFormer [2] model, which has State-Of-The-Art (SOTA) performance in optical flow estimation, as the teacher model. We then expect to create a more accurate optical flow by passing the obtained optical flow through U-Net [3] multiple times. In the experiments, we first create a teacher model and estimate the optical flow. Then, we use the estimated optical flow from the teacher model as the pseudo Ground Truth (GT) and train the student model. As a result of the experiments, we compared the accuracy of the proposed method and RIFE, but contrary to our expectations, the accuracy of the proposed method was lower than that of RIFE. This indicates that simply enlarging the teacher model does not necessarily lead to an improvement in the student model’s accuracy. From these results, it has become clear that further ingenuity is required in the video frame interpolation task.","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"4","bibliographic_titles":[{"bibliographic_title":"研究報告コンピュータビジョンとイメージメディア（CVIM）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2023-05-11","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"6","bibliographicVolumeNumber":"2023-CVIM-234"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"updated":"2025-01-19T12:37:57.451882+00:00","created":"2025-01-19T01:25:26.560953+00:00","id":225950}