Fast Vehicle and Pedestrian Detection Using Improved Mask R-CNN

Table 3

Network structure contrast of Resnet-50/Resnet-86/Resnet-101.

Backbone	Output size	Resnet-50	Resnet-86	Resnet-101
Conv_1	512 ∗ 512	7 ∗ 7, 64, stride2
Conv_2	256 ∗ 256	3 ∗ 3 maxpool, stride2
Conv_3	128 ∗ 128
Conv_4	64 ∗ 64
Conv_5	32 ∗ 32
	1 ∗ 1	Average pool, 1000-d fc, softmax