Table 2.

Detection results on the PASCAL visual object classes (VOC) 2007 test set (GPU: graphics processing unit, mAP: mean average precision, FPS: frames per second).

Model	Data	Backbone Network	Input Size	GPU	Framework	#Parameters	mAP	FPS
SSD300*	07 + 12	VGG	$300\times 300$	2080Ti	PyTorch	26.3 M	77.2	30
SSD300* 1	07 + 12	VGG	$300\times 300$	Titan X	Caffe	26.3 M	77.2	46
DSSD321	07 + 12	ResNet-101	$321\times 321$	Titan X	Caffe	- 2	78.6	9.5
SSD300-TSEFFM	07 + 12	VGG	$300\times 300$	2080Ti	PyTorch	-	78.6	7
DF-SSD300	07 + 12	DenseNet-S-32-1	$300\times 300$	Titan X	Caffe	15.2 M	78.9	11.6
RefineDet320	07 + 12	VGG	$320\times 320$	Titan X	Caffe	-	80.0	40.3
RefineDet320++	07 + 12	VGG	$320\times 320$	Titan X	PyTorch	-	81.1	27.8
SSD300-EMB	07 + 12	VGG	$300\times 300$	2080Ti	PyTorch	30.6 M	78.4	30

¹ SSD300 was tested with the PyTorch deep learning framework and RTX 2080Ti GPU, while SSD300* was tested with the Caffe deep learning framework and Titan X GPU. ² All data not mentioned in their papers are marked with ‘-’.