Pixel-Level Recognition of Pavement Distresses Based on U-Net

Table 2

The architecture of ResNet-34 and ResNet-50.

Layer name	34-layer	50-layer
Conv1	[7 × 7, 64]^a, stride 2
Max pool	[3 × 3], stride 2
Conv2_x
Conv3_x
Conv4_x
Conv5_x
Adaptive average pool	Output_size = 1
Adaptive max pool	Output_size = 1
Fully connected layer	Dimensionality = 7

^aThe number of the convolutional filters is 64, each of size 7 × 7.