Multiscale Efficient Channel Attention for Fusion Lane Line Segmentation

Table 5

The performance of using different pretrained parameters in fusion. F-MS denotes fusion only with multiscale strategy, and F-ECA denotes that fusion only with ECA.

Backbone	Method	REC	PRE	F2	Acc	FPS
	Fusion	92.24	48.57	77.12	98.44	79.0
ResNet18	F-MS	92.17	50.52	77.97	98.41	66.5
	F-ECA	92.13	50.66	78.03	98.57	78.3
	F-MS-ECA	93.08	51.21	78.34	98.62	60.2
	Fusion	92.75	49.37	78.23	98.54	75.3
ResNet34	F-MS	92.08	50.77	78.47	98.61	65.2
	F-ECA	92.12	50.36	78.46	98.61	73.9
	F-MS-ECA	92.23	51.38	78.57	98.65	59.5
	Fusion	93.09	49.84	78.53	98.58	65.9
ResNet50	F-MS	92.35	53.11	79.15	98.73	57.1
	F-ECA	92.27	52.97	78.83	98.69	62.8
	F-MS-ECA	93.17	54.37	79.72	98.81	53.6