[Retracted] DARSegNet: A Real-Time Semantic Segmentation Method Based on Dual Attention Fusion Module and Encoder-Decoder Network

Table 4

Comparison of accuracy, speed, and parameters of the lightweight segmentation model on the CamVid test set.

Method	Pretraining	GPU	MIoU (%)	Speed (FPS)
SegNet	ImageNet	TitanX	55.6	15
ENet	No	TitanX	51.3	61.2
BiseNet	ImageNet	1080Ti	65.6	175
ICNet	ImageNet	TitanX	67.1	34.5
SFNet (DF2)	No	1080Ti	70.4	134
CAS	No	Titan Xp	71.2	169
BiSeNet V2	No	1080Ti	68.7	116
BiSeNet V2	No	1080Ti	73.2	32.7
SwiftNet	No	1080Ti	72.6	75.9
DARSegNet (ours)	No	1080Ti	73.27	109
DARSegNet (ours)	ImageNet	1080Ti	73.96	109

The best results in each class are shown in bold.