Impostor Resilient Multimodal Metric Learning for Person Reidentification

Table 3

Top matching comparison on CUHK03.

	Labelled, = 100
	Method	= 1	= 5	= 10
F	LOMO [11]	52.2	82.23	92.14
	GoG [25]	67.3	91.0	96.0
DF	DCAF [41]	74.21	94.33	97.54
DFM	DGD [28]	75.3	-	-
	Quadruplet-net [33]	75.53	95.15	99.16
	MuDeep [30]	76.87	96.12	98.41
	E2E-CAN [31]	77.6	95.2	99.3
	JLML [34]	83.2	98.0	99.4
	DLPA [32]	85.4	97.6	99.4
M	SS-SVM [16]	57.0	85.7	94.3
	Null Sp. [42]	62.55	90.05	94.80
	SSM [43]	76.6	94.6	98.0
	WARCA [36]	78.38	94.55	-
	IRM3 only	78.83	95.97	98.37
	IRM3 + GVI ( = 15)	83.32	98.70	99.54
	IRM3 + CVI ( = 15)	86.17	99.02	99.68
	Detected, = 100
	Method
F	LOMO [11]	46.25	78.9	88.55
	GoG [25]	65.5	88.4	93.7
DF	SIR-CIR [26]	52.17	83.7	90.4
	DCAF [41]	67.99	91.04	95.36
DFM	LSTM [29]	57.3	80.10	88.3
	GS-CNN [27]	68.1	88.1	94.6
	E2E-CAN [31]	69.2	88.5	94.1
	MuDeep [30]	75.64	94.36	97.46
	JLML [34]	80.6	96.9	98.7
	DLPA [32]	81.6	97.3	98.4
M	L-1 graph [37]	39.0	-	-
	SS-SVM [16]	51.2	80.8	89.6
	Null Sp. [42]	54.70	84.75	94.80
	SSM [43]	72.7	92.4	96.1
	IRM3 only	72.98	91.7	93.02
	IRM3 + GVI ( = 15)	78.68	95.60	98.09
	IRM3 + CVI ( = 15)	80.77	96.94	98.67