IIPA-Net: Joint Illumination-Invariant and Pose-Aligned Feature Learning for Person Reidentification

Table 2

Experiment results of our framework compared to other state-of-the-art methods.

Dataset	Method	Rank-1	mAP
Market1501	IDE [41]	85.3	68.5
	Baseline [13]	92.8	89.4
	PCB [26]	92.3	77.4
	MHN-6 [3]; DSA [38]	95.1; 95.7	85.0; 87.6
	FlipReID [39]	95.8	94.7
	st-ReID [40]	98.0	95.5
	IIPA-Net	96.2	90.3
DukeMTMC-ReID	IDE [41]	73.2	52.8
	Baseline [13]	80.7	68.0
	PCB [26]	81.7	66.1
	MHN-6 [3]; DSA [38]	89.1; 86.2	77.2; 74.3
	FlipReID [39]	93.0	90.7
	st-ReID [40]	94.5	92.7
	IIPA-Net	90.8	83.3
Low-light Market	Baseline [13]	33.4	14.1
	Baseline+MSRCP; PCB [26]	49.4; 48.5	15.7; 16.2
	IIPA-Net	60.5	27.7
Low-light Duke	Baseline [13]	36.2	12.4
	Baseline+MSRCP; PCB [26]	40.4; 48.4	18.3; 21.0
	IIPA-Net	51.6	24.3

Bold: best results.