An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network

Multi-person pose estimation has been gaining considerable interest due to its use in several real-world applications, such as activity recognition, motion capture, and augmented reality. Although the improvement of the accuracy and speed of multi-person pose estimation techniques has been recently...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Changhyun Park, Hean Sung Lee, Woo Jin Kim, Han Byeol Bae, Jaeho Lee, Sangyoun Lee
Formato:	article
Lenguaje:	EN
Publicado:	MDPI AG 2021
Materias:	pose estimation convolutional neural network lightweight knowledge distillation Chemical technology TP1-1185
Acceso en línea:	https://doaj.org/article/5b7260bb507a4836b4ee492f48aad005
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:5b7260bb507a4836b4ee492f48aad005
record_format	dspace
spelling	oai:doaj.org-article:5b7260bb507a4836b4ee492f48aad0052021-11-25T18:58:08ZAn Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network10.3390/s212276401424-8220https://doaj.org/article/5b7260bb507a4836b4ee492f48aad0052021-11-01T00:00:00Zhttps://www.mdpi.com/1424-8220/21/22/7640https://doaj.org/toc/1424-8220Multi-person pose estimation has been gaining considerable interest due to its use in several real-world applications, such as activity recognition, motion capture, and augmented reality. Although the improvement of the accuracy and speed of multi-person pose estimation techniques has been recently studied, limitations still exist in balancing these two aspects. In this paper, a novel knowledge distilled lightweight top-down pose network (KDLPN) is proposed that balances computational complexity and accuracy. For the first time in multi-person pose estimation, a network that reduces computational complexity by applying a “Pelee” structure and shuffles pixels in the dense upsampling convolution layer to reduce the number of channels is presented. Furthermore, to prevent performance degradation because of the reduced computational complexity, knowledge distillation is applied to establish the pose estimation network as a teacher network. The method performance is evaluated on the MSCOCO dataset. Experimental results demonstrate that our KDLPN network significantly reduces 95% of the parameters required by state-of-the-art methods with minimal performance degradation. Moreover, our method is compared with other pose estimation methods to substantiate the importance of computational complexity reduction and its effectiveness.Changhyun ParkHean Sung LeeWoo Jin KimHan Byeol BaeJaeho LeeSangyoun LeeMDPI AGarticlepose estimationconvolutional neural networklightweightknowledge distillationChemical technologyTP1-1185ENSensors, Vol 21, Iss 7640, p 7640 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	pose estimation convolutional neural network lightweight knowledge distillation Chemical technology TP1-1185
spellingShingle	pose estimation convolutional neural network lightweight knowledge distillation Chemical technology TP1-1185 Changhyun Park Hean Sung Lee Woo Jin Kim Han Byeol Bae Jaeho Lee Sangyoun Lee An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
description	Multi-person pose estimation has been gaining considerable interest due to its use in several real-world applications, such as activity recognition, motion capture, and augmented reality. Although the improvement of the accuracy and speed of multi-person pose estimation techniques has been recently studied, limitations still exist in balancing these two aspects. In this paper, a novel knowledge distilled lightweight top-down pose network (KDLPN) is proposed that balances computational complexity and accuracy. For the first time in multi-person pose estimation, a network that reduces computational complexity by applying a “Pelee” structure and shuffles pixels in the dense upsampling convolution layer to reduce the number of channels is presented. Furthermore, to prevent performance degradation because of the reduced computational complexity, knowledge distillation is applied to establish the pose estimation network as a teacher network. The method performance is evaluated on the MSCOCO dataset. Experimental results demonstrate that our KDLPN network significantly reduces 95% of the parameters required by state-of-the-art methods with minimal performance degradation. Moreover, our method is compared with other pose estimation methods to substantiate the importance of computational complexity reduction and its effectiveness.
format	article
author	Changhyun Park Hean Sung Lee Woo Jin Kim Han Byeol Bae Jaeho Lee Sangyoun Lee
author_facet	Changhyun Park Hean Sung Lee Woo Jin Kim Han Byeol Bae Jaeho Lee Sangyoun Lee
author_sort	Changhyun Park
title	An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
title_short	An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
title_full	An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
title_fullStr	An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
title_full_unstemmed	An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network
title_sort	efficient approach using knowledge distillation methods to stabilize performance in a lightweight top-down posture estimation network
publisher	MDPI AG
publishDate	2021
url	https://doaj.org/article/5b7260bb507a4836b4ee492f48aad005
work_keys_str_mv	AT changhyunpark anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT heansunglee anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT woojinkim anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT hanbyeolbae anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT jaeholee anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT sangyounlee anefficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT changhyunpark efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT heansunglee efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT woojinkim efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT hanbyeolbae efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT jaeholee efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork AT sangyounlee efficientapproachusingknowledgedistillationmethodstostabilizeperformanceinalightweighttopdownpostureestimationnetwork
_version_	1718410464142032896

An Efficient Approach Using Knowledge Distillation Methods to Stabilize Performance in a Lightweight Top-Down Posture Estimation Network

Ejemplares similares