Table 2.

Comparison of our deep-learning architecture with other methods on the same dataset, where Faster R-CNN means faster region-based convolutional neural network, and SSD means single shot detector.

Method	Representation	mAP	training time
Faster R-CNN	ZF	0.764	30 min
Faster R-CNN	VGG16	0.847	40 min
Faster R-CNN	VGG_CNN_M_1024	0.754	35 min
SSD23	VGG16	0.874	4 h
Ours	ZF + RPN	0.885	10 min