Table 5 Comparison results of the model with and without the SepLDT block on the TuSimple and CULane datasets.

From: Aggregate global features into separable hierarchical lane detection transformer

Dataset

Method

Accuracy (%)

Para.(M)

Training time (h)

FPS

TuSimple

SepLDT Block

96.42

4.138

21.24

78

Non-SepLDT Block

92.50

7.144

46.36

64

CULane

SepLDT Block

72.89

4.138

61.75

75

Non-SepLDT Block

70.48

7.144

131.87

61

  1. Significant are in value [bold].
Back to article page