Designing Compact Convolutional Filters for Lightweight Human Pose Estimation

Table 5

Inference speed comparisons on the MSCOCO validation set. Speed refers to the result on non-GPU device. Speed refers to the result on GPU device. Bold values are the optimal results.

Method	Backbone	#Params	GFLOPs	Input size	AP	Speed	Speed
HRNet	HRNetV1-W32	28.5M	7.1		74.4	7.5	19.2
HRNet	HRNetV1-W32	28.5M	16		75.8	4	18.8
SimpleBaseline	ResNet-50	34.0M	8.9		70.4	8.1	273.1
NLite-HRNet-18	HRNet-W16	0.7M	0.19		62.8	11	18.9
WNLite-HRNet-18	HRNet-W16	1.3M	0.3		66	12	18.6
ShuffleNetV2 1×	ShuffleNetV2	7.6M	1.28		59.9	17	71.3
ShuffleNetV2 1×	ShuffleNetV2	7.6M	2.87		63.6	10	64.1
MobileNetV2 1×	MobileNetV2	9.6M	1.48		64.6	6.8	83.1
MobileNetV2 1×	MobileNetV2	9.6M	3.33		67.3	4.5	73.1
Lite-HRNet	Lite-HRNet-18	1.1M	0.2		64.8	12	17.4
Lite-HRNet	Lite-HRNet-18	1.1M	0.45		67.6	7.1	16.3
MobilePoseNet	MobileNetV3	1.5M	0.55		66.2	7.8	54.8
MobilePoseNet	MobileNetV3	1.5M	1.23		69.0	5.1	50.8