EURASIP Journal on Image and Video Processing

Table 4 Evaluation on the 10-class test sets of SegSemOutex and SegSemOutex-Synth

From: Semantic segmentation of textured mosaics

Metric	Class	Case
		N-N\(^{1}\)	S-S-Ind\(^{2}\)	S-S-All\(^{3}\)	S-N-Ind\(^{4}\)	S-N-All\(^{5}\)
PA	All	0.9940	0.9955 ± 0.0002	0.9976	0.9938 ± 0.0006	0.9969
mPA	All	0.9939	0.9954 ± 0.0002	0.9976	0.9937 ± 0.0006	0.9969
mIoU	All	0.9880	0.9909 ± 0.0005	0.9951	0.9874 ± 0.0012	0.9938
\(\text {IoU}_1\)	barleyrice002	0.9867	0.9889 ± 0.0012	0.9946	0.9783 ± 0.0056	0.9923
\(\text {IoU}_2\)	canvas005	0.9894	0.9912 ± 0.0003	0.9951	0.9892 ± 0.0002	0.9939
\(\text {IoU}_3\)	canvas022	0.9906	0.9925 ± 0.0003	0.9958	0.9907 ± 0.0003	0.9951
\(\text {IoU}_4\)	carpet005	0.9871	0.9896 ± 0.0011	0.9952	0.9812 ± 0.0051	0.9935
\(\text {IoU}_5\)	chips010	0.9846	0.9896 ± 0.0006	0.9942	0.9857 ± 0.0006	0.9920
\(\text {IoU}_6\)	paper004	0.9905	0.9924 ± 0.0005	0.9952	0.9914 ± 0.0003	0.9949
\(\text {IoU}_7\)	pasta006	0.9851	0.9897 ± 0.0006	0.9946	0.9874 ± 0.0005	0.9930
\(\text {IoU}_8\)	plastic005	0.9847	0.9896 ± 0.0003	0.9943	0.9871 ± 0.0002	0.9929
\(\text {IoU}_9\)	seeds004	0.9875	0.9905 ± 0.0003	0.9948	0.9893 ± 0.0002	0.9938
\(\text {IoU}_{10}\)	wallpaper006	0.9936	0.9949 ± 0.0002	0.9974	0.9939 ± 0.0002	0.9968

Best metric values shown in bold
\(^{1}\): N-N: model trained and tested on natural (SemSegOutex) textures
\(^{2}\): S-S-Ind: average and standard deviation of five seeded models trained and tested on synthetic (SemSegOutex-Synth) textures, one seed per model
\(^{3}\): S-S-All: single model trained and tested on synthetic (SemSegOutex-Synth) textures, all seeds combined
\(^{4}\): S-N-Ind: average and standard deviation of five seeded models trained on synthetic (SemSegOutex-Synth) textures (one seed per model) and tested on natural (SemSegOutex) textures
\(^{5}\): S-N-All: single model trained on synthetic (SemSegOutex-Synth) textures (all seeds combined) and tested on natural (SemSegOutex) textures

Back to article page