Weakly Supervised Deep Semantic Segmentation Using CNN and ELM with Semantic Candidate Regions

Table 2

Results on MSRC (mIoU in %) for weakly-supervised semantic segmentation with per-image labels.

Methods	building	grass	tree	cow	sheep	sky	aeroplane	water	face	car	bicycle	flower	sign	bird	book	chair	road	cat	dog	body	boat	average
MIM [17]	43	70	59	84	75	56	65	57	70	60	37	56	33	68	42	56	65	59	57	45	22	56
SR [18]	53	63	68	74	61	67	64	65	52	64	60	47	53	60	58	67	57	47	62	57	13	58
STC [19]	60	65	73	65	57	54	80	47	65	73	53	68	61	43	60	65	62	56	67	59	11	59
Ours	53	80	65	81	63	76	70	59	68	75	48	60	70	56	65	49	58	60	48	59	14	61