[Retracted] Multiobject Detection Algorithm Based on Adaptive Default Box Mechanism

Table 2

Results based on the PASCAL VOC 2012 test set.

Algorithm	Training set	Pretraining	Basic network	FPS	GPU	Input image	mAP (%)	mAR (%)
Faster RCNN [18]	VOC 2012	Yes	VGGNet	9	Titan X	∼600 × 1000	73.2	61.3
Faster RCNN [18]	VOC 2012	Yes	ResNet-101	3	Titan X	∼600 × 1000	76.4	63.5
YOLO v2 [23]	VOC 2012	Yes	Darknet-19	81	Titan X	352 × 352	73.7	61.7
SSD300−S0 [26]	VOC 2012	No	VGGNet	56	1080 Ti	300 × 300	69.6	58.2
SSD300 [26]	VOC 2012	Yes	VGGNet	56	1080 Ti	300 × 300	77.2	68.3
SSD300+coco [26]	VOC 2012 + COCO	Yes	VGGNet	56	1080 Ti	300 × 300	81.2	71.6
SSD512 [22]	VOC 2012	Yes	VGGNet	38	1080 Ti	512 × 512	78.5	69.4
SSD512+coco [26]	VOC 2012 + COCO	Yes	VGGNet	38	1080 Ti	512 × 512	83.2	72.5
DSSD321 [27]	VOC 2012	Yes	ResNet-101	12	1080 Ti	321 × 321	78.6	69.7
DSSD513 [27]	VOC 2012	Yes	ResNet-101	7	1080 Ti	513 × 513	81.5	72.4
RSSD300 [28]	VOC 2012	Yes	VGGNet	41	1080 Ti	300 × 300	78.5	68.9
RSSD512 [28]	VOC 2012	Yes	VGGNet	23	1080 Ti	512 × 512	80.8	71.2
Our SSD300−S0	VOC 2012	No	VGGNet	47	1080 Ti	300 × 300	73.2	62.4
Our SSD300	VOC 2012	Yes	VGGNet	47	1080 Ti	300 × 300	79.8	71.2
Our SSD300+coco	VOC 2012 + COCO	Yes	VGGNet	47	1080 Ti	300 × 300	83.4	74.1
Our SSD512	VOC 2012	Yes	VGGNet	30	1080 Ti	512 × 512	82.4	73.6
Our SSD512+coco	VOC 2012 + COCO	Yes	VGGNet	30	1080 Ti	512 × 512	85.6	75.9