Impostor Resilient Multimodal Metric Learning for Person Reidentification

Table 1

Top matching comparison on VIPeR.

	Single-shot, P = 316
	Method	= 1	= 5	= 10
F	LOMO [11]	40.0	-	80.51
	[24]	48.0	76.8	85.4
	GoG [25]	49.7	79.7	88.7
DF	SIR-CIR [26]	35.76	68.38	82.9
DMN	GS-CNN [27]	37.8	66.9	77.4
	DGD [28]	38.6	-	-
	LSTM [29]	42.4	68.7	79.4
	MuDeep [30]	43.03	74.36	85.76
	E2E-CAN [31]	47.2	79.2	89.2
	DLPA [32]	48.7	74.7	85.1
	Quadruplet-net [33]	49.05	73.10	81.96
	JLML [34]	50.2	74.2	84.3
M	ITL [35]	15.2	34.2	45.9
	LAFT [23]	29.6	-	69.3
	WARCA [36]	37.47	70.78	-
	LISTEN [21]	39.62	69.97	82.9
	L-1 graph [37]	41.5	-	-
	SS-SVM [16]	42.66	70.1	84.27
	SCSP [38]	53.54	82.59	91.49
	IRM3 only	45.92	82.90	91.63
	IRM3 + GVI ( = 15)	50.39	85.79	95.73
	IRM3 + CVI ( = 15)	52.81	87.95	97.29