Spectral Normalized CycleGAN with Application in Semisupervised Semantic Segmentation of Sonar Images

Table 4

Semisupervised semantic segmentation performance of different stabilization methods applied to CycleGAN on SCTDII datasets.

Method	The proportion of labels used
	1/2	2/5	3/10	1/5	1/10
None	0.63	0.61	0.41	0.32	0.27
EB-GAN [31]	0.67	0.62	0.47	0.39	0.32
MRP [33]	0.64	0.61	0.58	0.44	0.37
Gradient penalty [32]	0.63	0.63	0.50	0.42	0.32
SN [38]	0.69	0.66	0.64	0.54	0.43