Table 11 Ablation study on computational cost.

From: SwinCLNet: a robust framework for brain tumor segmentation via shifted window attention and cross-scale fusion

Component

Parameters (M)

GPU Memory (GB)

FLOPs (G)

Inference (ms)

3D-U-Net (Baseline)

19.0

1.75

98.5

118

+ RLKA

+0.36

+0.12

+8.8

+4

+ CSDualFusion

+4.40

+0.45

+2.3

+19

+ W-MSA/SW-MSA

+8.24

+0.49

+43.4

+36

SwinCLNet (Proposed)

32.0

2.81

153.0

177

Back to article page