EURASIP Journal on Image and Video Processing

Table 3 Influence of the number of regions (2, 3, 4, or 5) in test mosaics on the performance of DeepLabv3+ in terms of mIoU

From: Semantic segmentation of textured mosaics

Dataset	Experiment	# Regions				Global trend
		2	3	4	5
SegSemOutex	5-class	0.9888	0.9870	0.9840	0.9801	\(\Downarrow\)
	10-class	0.9911	0.9890	0.9860	0.9828	\(\Downarrow\)
SegSemDTD	5-class	0.9912	0.9901	0.9872	0.9846	\(\Downarrow\)
	10-class	0.9869	0.9889	0.9839	0.9796	\(\Downarrow\)
	5-class (hard)	0.9568	0.9474	0.9237	0.9100	\(\Downarrow\)
	10-class (hard)	0.7905	0.7747	0.7527	0.7471	\(\Downarrow\)

Back to article page