{"links":{},"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00234138","sets":["1164:4619:11539:11642"]},"path":["11642"],"owner":"44499","recid":"234138","title":["行動ラベルを用いた対照学習におけるアンカーの密度ベース選択による人間軌跡予測"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-05-08"},"_buckets":{"deposit":"0b60ec73-8c16-4ff7-a087-7e9675107f9b"},"_deposit":{"id":"234138","pid":{"type":"depid","value":"234138","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"行動ラベルを用いた対照学習におけるアンカーの密度ベース選択による人間軌跡予測","author_link":["637479","637478","637481","637480"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"行動ラベルを用いた対照学習におけるアンカーの密度ベース選択による人間軌跡予測"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"卒論スポットライトセッション (CVIM)","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2024-05-08","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"},{"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"},{"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/234138/files/IPSJ-CVIM24238007.pdf","label":"IPSJ-CVIM24238007.pdf"},"date":[{"dateType":"Available","dateValue":"2026-05-08"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-CVIM24238007.pdf","filesize":[{"value":"2.4 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":"20"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"6aa002d5-4e3e-4b3e-8304-c3ef802d8a4a","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"浦野, 耀太"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"武次, 広夢"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"前田, 孝泰"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"浮田, 宗伯"}],"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":"一人称視点からの人間軌跡予測は，人間と共存するロボットの衝突回避や動作支援に必要である．しかし，障害物により予測対象の得られる過去軌跡が少ないとき，予測が難しい．この問題に対処するため，各個人の行動を名付けた行動ラベルを対照学習に用いる．我々は，軌跡予測精度向上のために行動ラベルを用いた対照学習におけるアンカーを適切に選択する手法を提案する．従来は，対照学習におけるアンカーをランダムに選択していた．しかし，外れ値がアンカーとして選択されたとき，よく学習された特徴量空間を大きく崩してしまい，軌跡予測精度が悪化する．そこで本研究では，アンカーをその行動ラベルが持つ一般的な特徴量に決定することにより，外れ値の選択を回避する．そのために，我々は（1）クラスタリングを用いる手法（ASC），（2）近傍距離を計算する手法（ASND）を提案する．いずれもそのラベルを持つサンプル集合の中で，密度の高い場所からアンカーを選択することで外れ値を回避する手法である．ASC はクラスタ中心をアンカーとし，ASND はサンプル集合内でk近傍の距離を計算することによりアンカーを決定する．実験では，評価指標としてADE，FDE を用いて軌跡予測精度向上を確認し，特徴量空間や軌跡の可視化結果を考察することで提案手法が対照学習において有効であることを示す．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"8","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":"2024-05-08","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"7","bibliographicVolumeNumber":"2024-CVIM-238"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"updated":"2025-01-19T09:53:22.928599+00:00","created":"2025-01-19T01:35:48.193611+00:00","id":234138}