A Novel Approach of Intelligent Computing for Multiperson Pose Estimation with Deep High Spatial Resolution and Multiscale Features

Table 1

Comparisons on the COCO val2017 dataset.

Method	Backbone	Input size	Deconv. layers	Deconv. kernel size	AP
a	ResNet-50		3	4	70.4
b	ResNet-101		3	4	71.4
c	ResNet-152		3	4	72.0
d	ResNet-50		3	2	70.1
e	ResNet-50		3	3	70.3
f	ResNet-50		3	4	72.7
g	HSRNet-59		2	4	71.4
h	HSRNet-110		2	4	71.8
i	HSRNet-161		2	4	72.6
j	HSRNet-59		2	2	70.9
k	HSRNet-59		2	3	71.2
l	HSRNet-59		2	4	73.2