LiteDEKR: End‐to‐end lite 2D human pose estimation network
Abstract The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on high‐resolution network has superior performance, there is still room for improvement in terms of speed and lightweight. Here, a LiteDEKR, a 2D pose estima...
Saved in:
Published in | IET image processing Vol. 17; no. 12; pp. 3392 - 3400 |
---|---|
Main Authors | , , , , , |
Format | Journal Article |
Language | English |
Published |
Wiley
01.10.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Abstract | Abstract
The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on high‐resolution network has superior performance, there is still room for improvement in terms of speed and lightweight. Here, a LiteDEKR, a 2D pose estimation method that combines lightweight and accuracy, is proposed by designing a lightweight network based on DEKR and constructing two scientifically valid loss functions. The method, constructs a multi‐instance bias regression loss that matches the true distribution of keypoint bias, improves the accuracy of bias regression, and constructs a keypoint similarity loss with the object keypoint similarity index of keypoints as the optimization objective to achieve end‐to‐end training of the network. In addition, this paper has designed a lightweight DEKR, using LitePose as the backbone network. With the optimization of the above two loss functions, LiteDEKR not only achieves lightweight but also has high accuracy. Comparative experiments on the COCO and CrowdPose datasets show that compared to the current state‐of‐the‐art Contextual Instance Decoupling, LiteDEKR achieves a similar accuracy with only 10% of its network complexity. It also shows better robustness to low‐resolution input images. |
---|---|
AbstractList | Abstract The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on high‐resolution network has superior performance, there is still room for improvement in terms of speed and lightweight. Here, a LiteDEKR, a 2D pose estimation method that combines lightweight and accuracy, is proposed by designing a lightweight network based on DEKR and constructing two scientifically valid loss functions. The method, constructs a multi‐instance bias regression loss that matches the true distribution of keypoint bias, improves the accuracy of bias regression, and constructs a keypoint similarity loss with the object keypoint similarity index of keypoints as the optimization objective to achieve end‐to‐end training of the network. In addition, this paper has designed a lightweight DEKR, using LitePose as the backbone network. With the optimization of the above two loss functions, LiteDEKR not only achieves lightweight but also has high accuracy. Comparative experiments on the COCO and CrowdPose datasets show that compared to the current state‐of‐the‐art Contextual Instance Decoupling, LiteDEKR achieves a similar accuracy with only 10% of its network complexity. It also shows better robustness to low‐resolution input images. Abstract The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on high‐resolution network has superior performance, there is still room for improvement in terms of speed and lightweight. Here, a LiteDEKR, a 2D pose estimation method that combines lightweight and accuracy, is proposed by designing a lightweight network based on DEKR and constructing two scientifically valid loss functions. The method, constructs a multi‐instance bias regression loss that matches the true distribution of keypoint bias, improves the accuracy of bias regression, and constructs a keypoint similarity loss with the object keypoint similarity index of keypoints as the optimization objective to achieve end‐to‐end training of the network. In addition, this paper has designed a lightweight DEKR, using LitePose as the backbone network. With the optimization of the above two loss functions, LiteDEKR not only achieves lightweight but also has high accuracy. Comparative experiments on the COCO and CrowdPose datasets show that compared to the current state‐of‐the‐art Contextual Instance Decoupling, LiteDEKR achieves a similar accuracy with only 10% of its network complexity. It also shows better robustness to low‐resolution input images. |
Author | Lv, Xueqiang Tian, Lianghai Han, Jing Chen, Yuzhong Cai, Zangtai Hao, Wei |
Author_xml | – sequence: 1 givenname: Xueqiang surname: Lv fullname: Lv, Xueqiang organization: Beijing Key Laboratory of Internet Culture Digital Dissemination Beijing Information Science and Technology University Beijing China, The State Key Laboratory of Tibetan Intelligent Information Processing and Application Xining China – sequence: 2 givenname: Wei orcidid: 0000-0002-2148-816X surname: Hao fullname: Hao, Wei organization: Beijing Key Laboratory of Measurement and Control of Mechanical and Electrical System Technology Beijing Information Science and Technology University Beijing China – sequence: 3 givenname: Lianghai surname: Tian fullname: Tian, Lianghai organization: Beijing Key Laboratory of Internet Culture Digital Dissemination Beijing Information Science and Technology University Beijing China – sequence: 4 givenname: Jing orcidid: 0000-0001-9200-4874 surname: Han fullname: Han, Jing organization: Beijing Key Laboratory of Internet Culture Digital Dissemination Beijing Information Science and Technology University Beijing China – sequence: 5 givenname: Yuzhong surname: Chen fullname: Chen, Yuzhong organization: The State Key Laboratory of Tibetan Intelligent Information Processing and Application Xining China – sequence: 6 givenname: Zangtai surname: Cai fullname: Cai, Zangtai organization: The State Key Laboratory of Tibetan Intelligent Information Processing and Application Xining China |
BookMark | eNo9kMFKAzEQhoNUsK1efII9C1szyWaz8SBIu2qxIIieQ7I70a3tpmQj4s1H8Bl9Eret9DL_MAMfP9-IDFrfIiHnQCdAM3XZbAKbACskHJEhSAGpynM5OOxCnZBR1y0pFYoWYkiuF03EWfnwdJWUbf37_RN9P7Ctk1X_SNgseftYmzbZ-A4T7GKzNrHxbdJi_PTh_ZQcO7Pq8Ow_x-Tltnye3qeLx7v59GaRVlyymHKoFLOImasLVYABI9FJLvOsolDnQhTInQQha2OV5QILqzJmJauozF3h-JjM99zam6XehL5G-NLeNHp38OFVmxCbaoUa0RrpbG04qEwwaYUEw1yFgjKeI_Ssiz2rCr7rAroDD6jeWtRbi3pnkf8BuvNn1Q |
CitedBy_id | crossref_primary_10_3390_electronics13010143 |
Cites_doi | 10.1109/CVPRW56347.2022.00297 10.1007/978-3-030-58548-8_28 10.1109/CVPR52688.2022.01079 10.1007/978-3-319-10602-1_48 10.1109/CVPR46437.2021.01306 10.1609/aaai.v35i4.16446 10.1109/TPAMI.2020.2983686 10.1007/978-3-030-01231-1_29 10.1007/978-3-642-24136-9_20 10.1109/CVPR52688.2022.01078 10.1109/CVPR.2018.00742 10.1007/s11554‐021‐01132‐9 10.1109/ACCESS.2021.3069102 10.1109/TMM.2022.3159111 10.1007/978-3-031-20068-7_6 10.24963/ijcai.2022/120 10.1145/3469213.3470264 10.1109/ICCV48922.2021.01084 10.1109/CVPR46437.2021.01030 10.1109/CVPR46437.2021.01444 10.1109/ICCV.2019.00140 10.1109/CVPR52688.2022.01278 10.1109/ICCV.2019.00705 10.1109/CVPR.2019.01112 10.1109/CVPR.2019.00584 10.1109/CVPR.2018.00542 10.1109/ICCV48922.2021.00986 10.1109/ICCV48922.2021.01112 10.1109/CVPR42600.2020.00543 |
ContentType | Journal Article |
DBID | AAYXX CITATION DOA |
DOI | 10.1049/ipr2.12871 |
DatabaseName | CrossRef Directory of Open Access Journals |
DatabaseTitle | CrossRef |
DatabaseTitleList | CrossRef |
Database_xml | – sequence: 1 dbid: DOA name: Directory of Open Access Journals url: https://www.doaj.org/ sourceTypes: Open Website |
DeliveryMethod | fulltext_linktorsrc |
Discipline | Applied Sciences |
EISSN | 1751-9667 |
EndPage | 3400 |
ExternalDocumentID | oai_doaj_org_article_eeba7fbda3194527b571a2fce50236e1 10_1049_ipr2_12871 |
GroupedDBID | .DC 0R~ 1OC 24P 29I 4.4 5GY 6IK 8FE 8FG 8VB AAHHS AAHJG AAJGR AAYXX ABJCF ABQXS ACCFJ ACESK ACGFS ACIWK ACXQS ADZOD AEEZP AENEX AEQDE AFKRA AIWBW AJBDE ALMA_UNASSIGNED_HOLDINGS ALUQN ARAPS AVUZU BENPR BGLVJ CCPQU CITATION CS3 DU5 EBS EJD ESX GROUPED_DOAJ HCIFZ HZ~ IAO IFIPE IPLJI ITC JAVBF K1G L6V LAI M43 M7S MCNEO MS~ O9- OCL OK1 P2P P62 PTHSS QWB RIE RNS ROL RUI S0W ZL0 |
ID | FETCH-LOGICAL-c372t-31c92bee4fd8981a1a7ef73764c01d6558e3f7157dab9b35e8b942b72c076f8f3 |
IEDL.DBID | DOA |
ISSN | 1751-9659 |
IngestDate | Fri Oct 04 13:12:49 EDT 2024 Thu Sep 26 16:57:49 EDT 2024 |
IsDoiOpenAccess | true |
IsOpenAccess | true |
IsPeerReviewed | true |
IsScholarly | true |
Issue | 12 |
Language | English |
LinkModel | DirectLink |
MergedId | FETCHMERGED-LOGICAL-c372t-31c92bee4fd8981a1a7ef73764c01d6558e3f7157dab9b35e8b942b72c076f8f3 |
ORCID | 0000-0002-2148-816X 0000-0001-9200-4874 |
OpenAccessLink | https://doaj.org/article/eeba7fbda3194527b571a2fce50236e1 |
PageCount | 9 |
ParticipantIDs | doaj_primary_oai_doaj_org_article_eeba7fbda3194527b571a2fce50236e1 crossref_primary_10_1049_ipr2_12871 |
PublicationCentury | 2000 |
PublicationDate | 2023-10-01 |
PublicationDateYYYYMMDD | 2023-10-01 |
PublicationDate_xml | – month: 10 year: 2023 text: 2023-10-01 day: 01 |
PublicationDecade | 2020 |
PublicationTitle | IET image processing |
PublicationYear | 2023 |
Publisher | Wiley |
Publisher_xml | – name: Wiley |
References | e_1_2_10_23_1 e_1_2_10_24_1 e_1_2_10_21_1 e_1_2_10_22_1 e_1_2_10_20_1 e_1_2_10_2_1 e_1_2_10_4_1 e_1_2_10_18_1 e_1_2_10_3_1 e_1_2_10_19_1 e_1_2_10_6_1 e_1_2_10_16_1 e_1_2_10_5_1 e_1_2_10_17_1 e_1_2_10_8_1 e_1_2_10_14_1 e_1_2_10_7_1 e_1_2_10_15_1 e_1_2_10_12_1 e_1_2_10_9_1 e_1_2_10_13_1 e_1_2_10_34_1 e_1_2_10_10_1 e_1_2_10_33_1 e_1_2_10_11_1 e_1_2_10_32_1 e_1_2_10_31_1 e_1_2_10_30_1 e_1_2_10_29_1 e_1_2_10_27_1 e_1_2_10_28_1 e_1_2_10_25_1 e_1_2_10_26_1 |
References_xml | – ident: e_1_2_10_5_1 – ident: e_1_2_10_27_1 doi: 10.1109/CVPRW56347.2022.00297 – ident: e_1_2_10_33_1 – ident: e_1_2_10_25_1 doi: 10.1007/978-3-030-58548-8_28 – ident: e_1_2_10_9_1 doi: 10.1109/CVPR52688.2022.01079 – ident: e_1_2_10_16_1 doi: 10.1007/978-3-319-10602-1_48 – ident: e_1_2_10_23_1 doi: 10.1109/CVPR46437.2021.01306 – ident: e_1_2_10_32_1 – ident: e_1_2_10_24_1 doi: 10.1609/aaai.v35i4.16446 – ident: e_1_2_10_12_1 doi: 10.1109/TPAMI.2020.2983686 – ident: e_1_2_10_18_1 doi: 10.1007/978-3-030-01231-1_29 – ident: e_1_2_10_2_1 doi: 10.1007/978-3-642-24136-9_20 – ident: e_1_2_10_10_1 doi: 10.1109/CVPR52688.2022.01078 – ident: e_1_2_10_17_1 doi: 10.1109/CVPR.2018.00742 – ident: e_1_2_10_29_1 doi: 10.1007/s11554‐021‐01132‐9 – ident: e_1_2_10_30_1 doi: 10.1109/ACCESS.2021.3069102 – ident: e_1_2_10_6_1 doi: 10.1109/TMM.2022.3159111 – ident: e_1_2_10_21_1 doi: 10.1007/978-3-031-20068-7_6 – ident: e_1_2_10_20_1 doi: 10.24963/ijcai.2022/120 – ident: e_1_2_10_26_1 doi: 10.1145/3469213.3470264 – ident: e_1_2_10_15_1 doi: 10.1109/ICCV48922.2021.01084 – ident: e_1_2_10_28_1 doi: 10.1109/CVPR46437.2021.01030 – ident: e_1_2_10_8_1 doi: 10.1109/CVPR46437.2021.01444 – ident: e_1_2_10_31_1 doi: 10.1109/ICCV.2019.00140 – ident: e_1_2_10_13_1 doi: 10.1109/CVPR52688.2022.01278 – ident: e_1_2_10_7_1 doi: 10.1109/ICCV.2019.00705 – ident: e_1_2_10_14_1 doi: 10.1109/CVPR.2019.01112 – ident: e_1_2_10_11_1 doi: 10.1109/CVPR.2019.00584 – ident: e_1_2_10_4_1 doi: 10.1109/CVPR.2018.00542 – ident: e_1_2_10_34_1 doi: 10.1109/ICCV48922.2021.00986 – ident: e_1_2_10_3_1 – ident: e_1_2_10_19_1 doi: 10.1109/ICCV48922.2021.01112 – ident: e_1_2_10_22_1 doi: 10.1109/CVPR42600.2020.00543 |
SSID | ssj0059085 |
Score | 2.3410673 |
Snippet | Abstract
The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on... Abstract The 2D human pose estimation plays an important role in human‐computer interaction and action recognition. Although the method based on... |
SourceID | doaj crossref |
SourceType | Open Website Aggregation Database |
StartPage | 3392 |
SubjectTerms | convolutional neural nets pose estimation |
Title | LiteDEKR: End‐to‐end lite 2D human pose estimation network |
URI | https://doaj.org/article/eeba7fbda3194527b571a2fce50236e1 |
Volume | 17 |
hasFullText | 1 |
inHoldings | 1 |
isFullTextHit | |
isPrint | |
link | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwrV07T8MwELZQJxbeiPKSJVhD41ccMyABbVWBYEBU6hbZ8VliSau27PwEfiO_hFycojKxsGSIIiv67ny-k---j5BL4TMWtPUJ3kkl0lidGO9sAqkzOkDKXI7zzk_P2WgsHyZqsib1hT1hkR44AtcDcFYH523tK1Jx7ZRmlocSFHKfQyx8mFoVUzEGo5C3akYhUUQ-U2ZFTCpN720251cMC4VfR9EaY39ztAx3yFabE9Lb-C-7ZAOqPbLd5oe03X2LfXKDDBf9wePLNR1U_uvjczmtH1B5ioPElPdpo7hHZ9MFUGTPiGOJtIqt3gdkPBy83o-SVv8gKYXmyzo8loY7ABl8bnJmmdUQdB0RZJkynymVgwiaKe2tM04oyJ2R3GlepjoLeRCHpFNNKzgilAWoCwMrkH5P-tQ6kKUTUlhWyjRI0SUXKyiKWaS5KJrraWkKBKxoAOuSO0Tp5wukpm5e1AYrWoMVfxns-D8WOSGbqPseu-pOSWc5f4ezOjtYuvPGEb4BZqC4vg |
link.rule.ids | 315,786,790,870,2115,27957,27958 |
linkProvider | Directory of Open Access Journals |
openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=LiteDEKR%3A+End%E2%80%90to%E2%80%90end+lite+2D+human+pose+estimation+network&rft.jtitle=IET+image+processing&rft.au=Lv%2C+Xueqiang&rft.au=Hao%2C+Wei&rft.au=Tian%2C+Lianghai&rft.au=Han%2C+Jing&rft.date=2023-10-01&rft.issn=1751-9659&rft.eissn=1751-9667&rft.volume=17&rft.issue=12&rft.spage=3392&rft.epage=3400&rft_id=info:doi/10.1049%2Fipr2.12871&rft.externalDBID=n%2Fa&rft.externalDocID=10_1049_ipr2_12871 |
thumbnail_l | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1751-9659&client=summon |
thumbnail_m | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1751-9659&client=summon |
thumbnail_s | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1751-9659&client=summon |