Multistage Polymerization Network for Multiperson Pose Estimation

Table 3

Results on the COCO test-dev dataset. “” denotes using ensembled models. AP50 and AP75 indicate that we set the threshold to 0.5 and 0.75, respectively. APM indicates that the size of the detected target in the image ranges from 322 to 922, and APL indicates that the target range is greater than 922.

Method	Backbone	mAP	AP50	AP75	APM	APL
CMU-pose [15]	—	61.8	84.9	67.5	57.1	68.2
Mask-RCNN [7]	ResNet-50-FPN	63.1	87.3	68.7	57.8	71.4
MSPN [38]	ResNet-50	71.5	90.2	79.2	68.2	77.6
Integral pose regression [39]	ResNet-101	67.8	88.2	74.8	63.9	74.0
G-RMI [11]	ResNet-101	68.5	87.1	75.5	65.8	73.3
RSN [3]	RSN-18	70.4	88.8	77.7	67.2	72.2
SimpleBaseline [13]	ResNet-50	71.3	89.9	78.9	68.3	77.4
Graph-PCNN [24]	HR32	71.5	89.0	79.0	68.4	77.6
[26]	ResNet-50	71.7	91.1	79.6	68.6	77.1
CSM [12]	ResNet-101	71.8	91.3	80.1	68.7	77.3
DARK [25]	HR48	71.9	89.1	79.6	69.2	78.0
CPN+ [5]	ResNet-inception	72.1	90.5	78.9	67.9	78.1
Ours	MPN-16	68.8	87.8	75.9	66.1	74.3
Ours	MPN-32	70.5	88.5	77.5	67.5	76.5
Ours		71.9	89.2	78.8	69.1	77.8
Ours		72.3	89.2	79.4	69.4	78.3