Circle-Based Ratio Loss for Person Reidentification

Table 1

The experiment results and comparisons with some state-of-the-art works for Market-1501 and DukeMTMC-reID datasets on mAP and Rank-1.

Methods	Market-1501		DukeMTMC-reID
	mAP	Rank-1	mAP	Rank-1
IDE [7]	46.00	72.54	—	—
SVDNet [40]	62.1	82.3	56.8	76.7
AACN [41]	66.87	85.90	59.25	76.84
TriNet [13]	69.14	84.92	—	—
DPFL [42]	72.6	88.6	60.6	79.2
GLAD [43]	73.9	89.9	—	—
HA-CNN [44]	75.7	91.2	63.8	80.5
DuATM [45]	76.62	91.42	64.58	81.82
Deep-person [39]	79.58	92.31	64.80	80.90
PCB [6]	77.4	92.3	66.1	81.8
PCB + RPP [6]	81.6	93.8	69.2	83.3-
GSRW [46]	82.5	92.7	66.4	80.7
Baseline	81.65	92.31	71.10	83.44
Baseline + ratio loss	82.14	92.43	71.25	84.16
Baseline + OHEM	82.44	92.73	71.61	84.16
Baseline + ratio loss + OHEM (ours)	83.12	92.64	71.66	84.34

The bold values indicate the best results of all the methods on each metric. They are beneficial to compare between our proposed method and the other methods.