Downstream Semantic Segmentation Model for Low-Level Surface Crack Detection

Table 4

Time analysis regarding three model architectures based on input images of 512  512 resolution.

Model	Training time	Inference		Remark
	per 300 epochs	Time per image	FPS
DeepCrack [28]	—	0.153 s	6	One Nvidia GeForce GTX TITAN-X GPU (6.69 TFLOPS in FP32)
Modified U-Net [6]	2 h 15 m	0.040 s	25	One Nvidia GeForce RTX 2080 GPU
				(10.07 TFLOPS in FP32)
DeepLabV3-ResNet101 (ours)	6 h	0.078 s	13	One Nvidia GeForce RTX 2080 GPU
				(10.07 TFLOPS in FP32)