Person Reidentification Model Based on Multiattention Modules and Multiscale Residuals

Table 2

Experimental results of this model compared with other Re-ID models.

Type	Method	Market-1501		DukeMTMC-reID
		mAP	Rank-1	mAP	Rank-1
Stripe-based	AlignedReID [13]	77.7	90.6	67.4	81.2
	PCB + RPP [15]	81.6	93.8	69.2	83.3
	BFE [29]	85.0	94.5	75.8	88.7
	MGN [16]	86.9	95.7	78.4	88.7
	Pyramid [17]	88.2	95.7	79.0	89.0
	LocalCNN [30]	87.4	95.9	-	-
GAN-based	PN-GAN [31]	72.6	89.4	53.2	73.6
	DG-Net [32]	86.0	94.8	74.8	86.6
Global Feature	SVDNet [33]	62.1	82.3	56.8	76.7
	IA-Net [9]	83.1	94.4	73.4	87.1
	OS-Net [7]	84.9	94.8	73.5	88.6
	BagOfTricks [21]	85.9	94.5	76.4	86.4
	AGW [34]	87.8	95.1	79.6	89.0
Attention-based	MHN [10]	85.0	95.1	77.2	89.1
	ABD-Net [12]	88.3	95.6	78.6	89.0
	RGA-SC [35]	88.1	95.8	74.9	85.1
	SONA [12]	88.8	95.7	78.3	89.5
	Robust-ReID [8]	89.7	96.2	80.3	89.8
	Ours	88.2	96.2	76.4	89.6
	Ours(Re-rank）	94.6	96.5	89.5	91.9