Deep Camera-Aware Metric Learning for Person Reidentification

Table 5

Ablation study results (%) on three datasets. If without dynamic weighting, the ratio of ID loss and CA loss is 1 : 1. ID is identification loss while CA is a camera-aware loss.

Dataset	Model	Rank-1	mAP
Market1501	ResNet50+ID	87.8	71.6
	ResNet50+ID+CA	91.6	77.9
	ResNet50+ID+CA+DW	92.7	78.7
DukeMTMC-reID	ResNet50+ID	80.2	62.2
	ResNet50+ID+CA	81.3	65.0
	ResNet50+ID+CA+DW	83.6	67.8
MSMT17	ResNet50+ID	45.8	29.9
	ResNet50+ID+CA	47.5	31.4
	ResNet50+ID+CA+DW	50.5	33.1