Table 2.

Comparative analysis of different models trained using our proposed CCS-based self-training technique with standard self-training in terms of average accuracy (Avg. Acc), mean intersection over union (mIoU), and dice similarity score.

Standard self-training						CCS-based self-training
Teacher model	Student model	Backbone	Avg. Acc	mIoU	Dice	Teacher model	Student model	Backbone	Avg. Acc	mIoU	Dice
Deeplabv3 (ResNet-101)	Deeplabv3	ResNet-50	91.77	45.07	0.45	Deeplabv3 (ResNet-101)	Deeplabv3	ResNet-50	94.57	46.35	0.46
		ResNet-101	93.22	47.83	0.47			ResNet-101	99.43	50.73	0.50
		Mobilenet-v3	90.15	44.38	0.44			Mobilenet-v3	95.08	45.60	0.45
	FCN	ResNet-50	90.04	44.81	0.44		FCN	ResNet-50	97.61	47.52	0.47
		ResNet-101	92.98	45.95	0.45			ResNet-101	95.84	45.87	0.45
	LRASPP	Mobilenet-v3	87.41	42.73	0.42		LRASPP	Mobilenet-v3	91.72	40.16	0.40

Significant values are in bold.