SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection

Feature Pyramid Network (FPN) is used as the neck of current popular object detection networks. Research has shown that the structure of FPN has some defects. In addition to the loss of information caused by the reduction of the channel number, the features scale of different levels are also differe...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Zhiqiang Zhang, Xin Qiu, Yongzhou Li
Formato:	article
Lenguaje:	EN
Publicado:	MDPI AG 2021
Materias:	object detection feature pyramid level imbalance correlation convolution Chemical technology TP1-1185
Acceso en línea:	https://doaj.org/article/e780f4eb6d264a7db5c5c866245b3988
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:e780f4eb6d264a7db5c5c866245b3988
record_format	dspace
spelling	oai:doaj.org-article:e780f4eb6d264a7db5c5c866245b39882021-11-11T19:08:08ZSEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection10.3390/s212171361424-8220https://doaj.org/article/e780f4eb6d264a7db5c5c866245b39882021-10-01T00:00:00Zhttps://www.mdpi.com/1424-8220/21/21/7136https://doaj.org/toc/1424-8220Feature Pyramid Network (FPN) is used as the neck of current popular object detection networks. Research has shown that the structure of FPN has some defects. In addition to the loss of information caused by the reduction of the channel number, the features scale of different levels are also different, and the corresponding information at different abstract levels are also different, resulting in a semantic gap between each level. We call the semantic gap level imbalance. Correlation convolution is a way to alleviate the imbalance between adjacent layers; however, how to alleviate imbalance between all levels is another problem. In this article, we propose a new simple but effective network structure called Scale-Equalizing Feature Pyramid Network (SEFPN), which generates multiple features of different scales by iteratively fusing the features of each level. SEFPN improves the overall performance of the network by balancing the semantic representation of each layer of features. The experimental results on the MS-COCO2017 dataset show that the integration of SEFPN as a standalone module into the one-stage network can further improve the performance of the detector, by ∼1AP, and improve the detection performance of Faster R-CNN, a typical two-stage network, especially for large object detection <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>A</mi><msub><mi>P</mi><mi>L</mi></msub></mrow></semantics></math></inline-formula>∼2AP.Zhiqiang ZhangXin QiuYongzhou LiMDPI AGarticleobject detectionfeature pyramidlevel imbalancecorrelation convolutionChemical technologyTP1-1185ENSensors, Vol 21, Iss 7136, p 7136 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	object detection feature pyramid level imbalance correlation convolution Chemical technology TP1-1185
spellingShingle	object detection feature pyramid level imbalance correlation convolution Chemical technology TP1-1185 Zhiqiang Zhang Xin Qiu Yongzhou Li SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
description	Feature Pyramid Network (FPN) is used as the neck of current popular object detection networks. Research has shown that the structure of FPN has some defects. In addition to the loss of information caused by the reduction of the channel number, the features scale of different levels are also different, and the corresponding information at different abstract levels are also different, resulting in a semantic gap between each level. We call the semantic gap level imbalance. Correlation convolution is a way to alleviate the imbalance between adjacent layers; however, how to alleviate imbalance between all levels is another problem. In this article, we propose a new simple but effective network structure called Scale-Equalizing Feature Pyramid Network (SEFPN), which generates multiple features of different scales by iteratively fusing the features of each level. SEFPN improves the overall performance of the network by balancing the semantic representation of each layer of features. The experimental results on the MS-COCO2017 dataset show that the integration of SEFPN as a standalone module into the one-stage network can further improve the performance of the detector, by ∼1AP, and improve the detection performance of Faster R-CNN, a typical two-stage network, especially for large object detection <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>A</mi><msub><mi>P</mi><mi>L</mi></msub></mrow></semantics></math></inline-formula>∼2AP.
format	article
author	Zhiqiang Zhang Xin Qiu Yongzhou Li
author_facet	Zhiqiang Zhang Xin Qiu Yongzhou Li
author_sort	Zhiqiang Zhang
title	SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
title_short	SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
title_full	SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
title_fullStr	SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
title_full_unstemmed	SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
title_sort	sefpn: scale-equalizing feature pyramid network for object detection
publisher	MDPI AG
publishDate	2021
url	https://doaj.org/article/e780f4eb6d264a7db5c5c866245b3988
work_keys_str_mv	AT zhiqiangzhang sefpnscaleequalizingfeaturepyramidnetworkforobjectdetection AT xinqiu sefpnscaleequalizingfeaturepyramidnetworkforobjectdetection AT yongzhouli sefpnscaleequalizingfeaturepyramidnetworkforobjectdetection
_version_	1718431617688535040

SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection

Ejemplares similares