Small Object Detection Network Based on Feature Information Enhancement

Table 1

Test results on PASCAL VOC 2007 test set.

Method	Backbone	Size	GPU	mAP (%)
YOLOv1 [8]	GoogleNet	448 × 448	Tian X	63.4
YOLOv2 [9]	Darknet-19	352 × 352	Tian X	73.7
SSD300 [7]	VGG-16	300 × 300	2080Ti	77.2
SSD512 [7]	VGG-16	512 × 512	2080Ti	79.8
RefineDet320 [41]	VGG-16	320 × 320	Tian X	80.0
RefineDet512 [41]	VGG-16	512 × 512	Tian X	81.8
RFB300 [17]	VGG-16	300 × 300	2080Ti	80.1
RFB512 [17]	VGG-16	512 × 512	2080Ti	81.2
AFP-SSD [33]	VGG-16	300 × 300	Tian X	79.3
RSSD300 [42]	VGG-16	300 × 300	Tian X	78.5
RSSD512 [42]	VGG-16	512 × 512	Tian X	80.8
FSSD300 [16]	VGG-16	300 × 300	1080Ti	78.8
FSSD512 [16]	VGG-16	512 × 512	1080Ti	80.9
SEFN300 [43]	VGG-16	300 × 300	1080Ti	79.6
SEFN512 [43]	VGG-16	512 × 512	1080Ti	81.2
FIEN300	VGG-16	300 × 300	2080Ti	80.2
FIEN512	VGG-16	512 × 512	2080Ti	81.3
FIEN_RFB300	VGG-16	300 × 300	2080Ti	80.5
FIEN_RFB512	VGG-16	512 × 512	2080Ti	81.6