Crossmodality Person Reidentification Based on Global and Local Alignment

Table 4

Comparison with different variants of CNNs on SYSU-MM01 dataset and our dataset under the all-search mode. The R1, R10, and R20 denote rank 1, rank 10, and rank 20 accuracies (%), respectively. The mAP denotes mean average precision score (%).

Dataset	CNNs	All-search
		Single-shot				Multishot
		R1	R10	R20	mAP	R1	R10	R20	mAP
SYSU-MM01	ResNet-50	38.1	80.7	89.9	36.9	45.1	85.7	93.8	29.5
	IBN-ResNet-50	38.8	81.5	89.9	36.5	48.4	88.0	94.1	29.6
	ResNext-50	37.9	81.0	89.7	36.0	47.1	86.9	94.3	29.1
	SE-ResNet-50	39.6	82.6	91.1	37.1	48.5	88.6	95.0	30.1
Our dataset	ResNet-50	41.2	83.6	91.5	38.9	50.7	89.2	95.2	31.8
	IBN-ResNet-50	40.3	81.8	90.3	37.5	50.6	88.0	94.1	30.5
	ResNext-50	40.9	83.1	90.9	38.7	49.8	88.5	94.9	31.6
	SE-ResNet-50	43.8	86.2	93.4	40.9	53.8	91.9	96.7	33.7