Table 6 The segmentation accuracy and... | Emerald Publishing

Table 6

The segmentation accuracy and efficiency of lightweight MOS methods on DAVIS 2017 dataset.

No.	Year	Method	GPU	J&F Mean (%)↑	J Mean (%)↑	F Mean (%)↑	Inference Speed (fps) ↑	# Param (M) ↓	Model Size (MB) ↓
1	2021	AOT-T [232]	Tesla V100	72	68.3	75.7	51.4	5.7	23.3
2	2020	LWL [13]		70.8	68.2	73.5	14	-	-
3	2021	RM Net [225]		83.5	81	86	12^*	53	202
4	2020	SAT-fast [32]	RTX 2080Ti	69.5	65.4	73.6	60	16	64
5	2020	SAT [32]		72.3	68.6	76	39	72	288
6	2019	PiWiVOS-F [152]	Titan Xp	54.9	55.7	54	85	8	31.3
7	2020	TVOS [247]		72.3	69.9	74.7	37	12.7	50
8	2019	RANet [219]		65.7	63.2	68.2	30*	61.5	246
9	2019	AGSS [112]		67.4	64.9	69.9	10	48	197
10	2019	DTN [245]	GTX 1080Ti	67.4	64.2	70.6	14.3^*	-	-
11	2020	Fasttmu [188]		70.6	69.1	72.1	11	5.2	37.8
12	2019	SiamMask(ResNet-50) [210]	RTX 2080	56.4	54.3	58.5	55	27	105.9
13	2021	SwiftNet (ResNet-18) [206]	Tesla P100	77.8	75.7	79.9	70	32.5	130
14	2021	SwiftNet (ResNet-50) [206]		81.1	78.3	83.9	25	37.4	149
15	2020	GC [105]	Tesla P40	71.4	69.3	73.5	25^*	45.9	175
16	2021	TTVOS(HRNet) [154]	-	62.1	-	-	78.3^*	1.61	6
17	2021	TTVOS(ResNet-50) [154]	-	69.5	-	-	37.7^*	14.8	159
18	2021	G-FRTM-fast (RN18, τ =1) [155]	-	71.7	-	-	37.6^*	-	-
19	2020	MSN [222]	-	74.1	71.4	76.8	10	-	-

No.	Year	Method	GPU	J&F Mean (%)↑	J Mean (%)↑	F Mean (%)↑	Inference Speed (fps) ↑	# Param (M) ↓	Model Size (MB) ↓
1	2021	AOT-T [232]	Tesla V100	72	68.3	75.7	51.4	5.7	23.3
2	2020	LWL [13]		70.8	68.2	73.5	14	-	-
3	2021	RM Net [225]		83.5	81	86	12^*	53	202
4	2020	SAT-fast [32]	RTX 2080Ti	69.5	65.4	73.6	60	16	64
5	2020	SAT [32]		72.3	68.6	76	39	72	288
6	2019	PiWiVOS-F [152]	Titan Xp	54.9	55.7	54	85	8	31.3
7	2020	TVOS [247]		72.3	69.9	74.7	37	12.7	50
8	2019	RANet [219]		65.7	63.2	68.2	30*	61.5	246
9	2019	AGSS [112]		67.4	64.9	69.9	10	48	197
10	2019	DTN [245]	GTX 1080Ti	67.4	64.2	70.6	14.3^*	-	-
11	2020	Fasttmu [188]		70.6	69.1	72.1	11	5.2	37.8
12	2019	SiamMask(ResNet-50) [210]	RTX 2080	56.4	54.3	58.5	55	27	105.9
13	2021	SwiftNet (ResNet-18) [206]	Tesla P100	77.8	75.7	79.9	70	32.5	130
14	2021	SwiftNet (ResNet-50) [206]		81.1	78.3	83.9	25	37.4	149
15	2020	GC [105]	Tesla P40	71.4	69.3	73.5	25^*	45.9	175
16	2021	TTVOS(HRNet) [154]	-	62.1	-	-	78.3^*	1.61	6
17	2021	TTVOS(ResNet-50) [154]	-	69.5	-	-	37.7^*	14.8	159
18	2021	G-FRTM-fast (RN18, τ =1) [155]	-	71.7	-	-	37.6^*	-	-
19	2020	MSN [222]	-	74.1	71.4	76.8	10	-	-

Note:^* The inference speeds of these models were evaluated on the DAVIS 2016 dataset in the original papers.