WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network

Human pose estimation is a task that involves locating the body joints in an image. Current deep learning models accurately estimate the locations of these joints. However, they struggle with smaller joints, such as the wrist and ankle, leading to lower accuracy. To address this problem, current mod...

Full description

Saved in:
Bibliographic Details
Main Authors: Esraa Samkari, Muhammad Arif, Manal AlGhamdi, Mohammed A. Al Ghamdi
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10707605/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850282432949911552
author Esraa Samkari
Muhammad Arif
Manal AlGhamdi
Mohammed A. Al Ghamdi
author_facet Esraa Samkari
Muhammad Arif
Manal AlGhamdi
Mohammed A. Al Ghamdi
author_sort Esraa Samkari
collection DOAJ
description Human pose estimation is a task that involves locating the body joints in an image. Current deep learning models accurately estimate the locations of these joints. However, they struggle with smaller joints, such as the wrist and ankle, leading to lower accuracy. To address this problem, current models add more layers and make the model deeper to achieve higher accuracy. However, this solution adds complexity to the model. Therefore, we present an efficient network that can estimate small joints by capturing more features by increasing the network’s channels. Our network structure follows multiple stages and multiple branches while maintaining high-resolution output along the network. Hence, we called this network Wide High-Resolution Network (WideHRNet). WideHRNet provides several advantages. First, it runs in parallel and provides a high-resolution output. Second, unlike heavyweight networks, WideHRNet obtains superior results using a few layers. Third, the complexity of WideHRNet can be controlled by adjusting the hyperparameter of expansion channels. Fourth, the performance of WideHRNet is further enhanced by adding the attention mechanism. Experimental results on the MPII dataset show that the WideHRNet outperforms state-of-the-art efficient models, achieving 88.47% with the attention block.
format Article
id doaj-art-d41a4e8ca65e4c879a37f5e40460d428
institution OA Journals
issn 2169-3536
language English
publishDate 2024-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-d41a4e8ca65e4c879a37f5e40460d4282025-08-20T01:47:58ZengIEEEIEEE Access2169-35362024-01-011214899014900010.1109/ACCESS.2024.347619610707605WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution NetworkEsraa Samkari0https://orcid.org/0009-0007-1413-0403Muhammad Arif1https://orcid.org/0000-0003-0513-9872Manal AlGhamdi2https://orcid.org/0000-0002-7895-6999Mohammed A. Al Ghamdi3https://orcid.org/0000-0002-5993-5236Department of Computer Science and Artificial Intelligence, Umm Al-Qura University, Makkah, Saudi ArabiaDepartment of Computer Science and Artificial Intelligence, Umm Al-Qura University, Makkah, Saudi ArabiaDepartment of Computer Science and Artificial Intelligence, Umm Al-Qura University, Makkah, Saudi ArabiaDepartment of Computer Science and Artificial Intelligence, Umm Al-Qura University, Makkah, Saudi ArabiaHuman pose estimation is a task that involves locating the body joints in an image. Current deep learning models accurately estimate the locations of these joints. However, they struggle with smaller joints, such as the wrist and ankle, leading to lower accuracy. To address this problem, current models add more layers and make the model deeper to achieve higher accuracy. However, this solution adds complexity to the model. Therefore, we present an efficient network that can estimate small joints by capturing more features by increasing the network’s channels. Our network structure follows multiple stages and multiple branches while maintaining high-resolution output along the network. Hence, we called this network Wide High-Resolution Network (WideHRNet). WideHRNet provides several advantages. First, it runs in parallel and provides a high-resolution output. Second, unlike heavyweight networks, WideHRNet obtains superior results using a few layers. Third, the complexity of WideHRNet can be controlled by adjusting the hyperparameter of expansion channels. Fourth, the performance of WideHRNet is further enhanced by adding the attention mechanism. Experimental results on the MPII dataset show that the WideHRNet outperforms state-of-the-art efficient models, achieving 88.47% with the attention block.https://ieeexplore.ieee.org/document/10707605/Convolution neural networkefficient networkhuman pose estimationwide network
spellingShingle Esraa Samkari
Muhammad Arif
Manal AlGhamdi
Mohammed A. Al Ghamdi
WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
IEEE Access
Convolution neural network
efficient network
human pose estimation
wide network
title WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
title_full WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
title_fullStr WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
title_full_unstemmed WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
title_short WideHRNet: An Efficient Model for Human Pose Estimation Using Wide Channels in Lightweight High-Resolution Network
title_sort widehrnet an efficient model for human pose estimation using wide channels in lightweight high resolution network
topic Convolution neural network
efficient network
human pose estimation
wide network
url https://ieeexplore.ieee.org/document/10707605/
work_keys_str_mv AT esraasamkari widehrnetanefficientmodelforhumanposeestimationusingwidechannelsinlightweighthighresolutionnetwork
AT muhammadarif widehrnetanefficientmodelforhumanposeestimationusingwidechannelsinlightweighthighresolutionnetwork
AT manalalghamdi widehrnetanefficientmodelforhumanposeestimationusingwidechannelsinlightweighthighresolutionnetwork
AT mohammedaalghamdi widehrnetanefficientmodelforhumanposeestimationusingwidechannelsinlightweighthighresolutionnetwork