Fast Vehicle and Pedestrian Detection Using Improved Mask R-CNN

Table 5

Experimental data.

Backbone_class	FPN + resnet101_81	FPN + resnet86_81	FPN + resnet50_81	FPN + resnet101_3	FPN + resnet86_3	FPN + resnet50_3
	59.11	58.74	46.16	75.69	74.84	67.09
	55.02	54.88	42.04	70.24	69.23	62.15
	47.45	47.88	35.13	60.34	58.77	51.98
	33.35	33.88	22.85	39.96	39.04	31.59
	76.76	75.96	70.60	80.10	78.89	73.21
	53.29	50.87	45.77	52.73	51.38	46.99
	83.15	81.26	75.86	84.09	83.28	76.71
Min_train_loss	0.9564	0.9092	1.287	0.6592	0.7138	0.8643

The evaluation standard uses the mAP. The number in the upper right corner of mAP indicates the number of input/output units (IOUs). The words “person,” “car,” and “bus” in the bottom right corner mean the detection of the single category. Min_train_loss refers to the loss value of each model after training.