Table 2.

Details of Deep Learning Meta-architectures and Feature Extractors.

Meta-Architecture	Feature Extractor	Bounding Box	Loss Function
Faster R-CNN	VGG-16 ResNet-50 ResNet-101 ResNet-152 ResNeXt-50	$\left[\frac{x_c}{w_a},\frac{y_c}{h_a},\log w,\log h\right]$	Smooth$L_1$
SSD	ResNet-50	$\left[\frac{x_c}{w_a},\frac{y_c}{h_a},\log w,\log h\right]$	Smooth$L_1$
R-FCN	ResNet-50	$\left[\frac{x_c}{w_a},\frac{y_c}{h_a},\log w,\log h\right]$	Smooth$L_1$

Faster R-CNN: faster region-based convolutional neural network; SSD: single shot detector; R-FCN: region-based fully convolutional network.