Designing Compact Convolutional Filters for Lightweight Human Pose Estimation

Table 2

Comparisons of results on the MSCOCO validation set.

Method	Backbone	Input	#Params	GFLOPs
8-stage hourglass [6]	Hourglass		25.6M	26.2	66.9	—	—	—	—	—
CPN [31]	ResNet-50		27.0M	6.2	68.4	—	—	—	—	—
SimpleBaseline [25]	ResNet-50		34.0M	8.9	70.4	88.6	78.3	67.1	77.2	76.3
HRNet-W32 [5]	ResNet-50		28.5M	12.4	73.4	89.5	80.7	70.2	80.1	79.8
DARK [32]	HRNetV1-W48		63.6M	3.6	71.9	89.1	79.6	69.2	78	77.9
MobileNetV2 [19]	MobileNetV2		9.6M	1.48	64.6	87.4	72.3	61.1	71.2	70.7
MobileNetV2 1×	MobileNetV2		9.6M	3.33	67.3	87.9	74.3	62.8	74.7	72.9
ShuffleNetV2 [33]	ShuffleNetV2		7.6M	1.28	59.9	85.4	66.3	56.6	66.2	66.4
ShuffleNetV2 1×	ShuffleNetV2		7.6M	2.87	63.6	86.5	70.5	59.5	70.7	69.7
Small HRNet	HRNet-W16		1.3M	0.54	55.2	83.7	62.4	52.3	61	62.1
Small HRNet	HRNet-W16		1.3M	1.21	56	83.8	63	52.4	62.6	62.6
Lite-HRNet	Lite-HRNet-18		1.1M	0.20	64.8	86.7	73	62.1	70.5	71.2
Lite-HRNet	Lite-HRNet-18		1.1M	0.45	67.6	87.8	75	64.5	73.7	73.7
MobilePoseNet	MobilNetV3		1.5M	0.55	66.2	87.3	74.2	63.1	72.5	72.4
MobilePoseNet	MobilNetV3		1.5M	1.23	69	88.2	75.9	65.5	75.5	74.9