M2YOLOF: Based on effective receptive fields and multiple-in-single-out encoder for object detection

doi:10.1016/j.eswa.2022.118928

	M2YOLOF: Based on effective receptive fields and multiple-in-single-out encoder for object detection
	Wang, Qijin 1,2,3,4,5; Qian, Yu 3; Hu, Yating 3; Wang, Chao 3; Ye, Xiaodong 1; Wang, Hongqiang 1,2
刊名	EXPERT SYSTEMS WITH APPLICATIONS
	2023-03-01
卷号	213
关键词	Object detection Deep learning YOLOF Effective receptive field
ISSN号	0957-4174
DOI	10.1016/j.eswa.2022.118928
通讯作者	Wang, Hongqiang(hqwang@ustc.edu)
英文摘要	Object detection under one-level feature is a difficult task, which requires that different scale object represen-tations can be extracted on one feature map, as well as the balance between quality and quantity of positive samples play a key role in model training. YOLOF with real-time detection speed solves the partial problems about object scale and sample quantity balance. To further improve performance especially in smaller objects, we propose a new object detector called M2YOLOF. The main ingredients are a multi-in-single-out encoder that joints attention to strengthen the local feature and global representation of each multi-scale object, and a dy-namic sample selection policy that using effective receptive fields to rationalize the quantity of positive samples. M2YOLOF strengthen the contextual details of feature map and balances the rationality of training samples. Extensive experiments on COCO benchmark prove the effectiveness of our method, with an image size of [1333,800], using ResNet50 as backbone, running at 29 FPS on 2080Ti and achieving 39.2 AP. It is 1.7 AP higher than YOLOF but GFLOPs of our method only increases by <9%.
资助项目	National Natural Science Foundation of China ; Academic funding project for top talents of disciplines in Colleges and universities of Anhui Province ; [61973295] ; [61773360] ; [201904a07020092] ; [gxbjZD2020096]
WOS研究方向	Computer Science ; Engineering ; Operations Research & Management Science
语种	英语
出版者	PERGAMON-ELSEVIER SCIENCE LTD
WOS记录号	WOS:000867522400002
资助机构	National Natural Science Foundation of China ; Academic funding project for top talents of disciplines in Colleges and universities of Anhui Province
内容类型	期刊论文
源URL	[http://ir.hfcas.ac.cn:8080/handle/334002/129428]
专题	中国科学院合肥物质科学研究院
通讯作者	Wang, Hongqiang
作者单位	1.Chinese Acad Sci, Inst Intelligent Machines, Hefei Inst Phys Sci, Hefei, Peoples R China 2.Chinese Acad Sci, Hefei Inst Phys Sci, Zhongqi AI Joint Lab, Hefei, Peoples R China 3.Anhui Jianzhu Univ, Sch Elect & Informat Engn, Hefei, Peoples R China 4.Anhui Xinhua Univ, Hefei, Peoples R China 5.Univ Sci & Technol China, Hefei, Peoples R China
推荐引用方式 GB/T 7714	Wang, Qijin,Qian, Yu,Hu, Yating,et al. M2YOLOF: Based on effective receptive fields and multiple-in-single-out encoder for object detection[J]. EXPERT SYSTEMS WITH APPLICATIONS,2023,213.
APA	Wang, Qijin,Qian, Yu,Hu, Yating,Wang, Chao,Ye, Xiaodong,&Wang, Hongqiang.(2023).M2YOLOF: Based on effective receptive fields and multiple-in-single-out encoder for object detection.EXPERT SYSTEMS WITH APPLICATIONS,213.
MLA	Wang, Qijin,et al."M2YOLOF: Based on effective receptive fields and multiple-in-single-out encoder for object detection".EXPERT SYSTEMS WITH APPLICATIONS 213(2023).