A Deep Neural Network Based on Circular Representation for Target Detection

Table 4

Comparison table between RebarDet and other mainstream object detection models in AB-score, inference speed, and parameter quantity.

Method	Backbone	Input size	AB-score	Inference speed (fps)	Params (M)
SSD	VGG-16		0.7233	3.5	33.51
RetinaNet [34]	ResNet-101		0.7536	3.7	55.37
Faster-RCNN	ResNet-101		0.7745	3.5	60.18
ATSS [35]	ResNet-101		0.7812	3.5	50.91
YOLOv5	yolov5x		0.7983	7.1	40.97
RebarDet (ours)	Hourglass-104		0.8114	6.9	35.24