Research Article

Designing Compact Convolutional Filters for Lightweight Human Pose Estimation

Table 4

Comparisons on the MPII val set. The GFLOPs is computed with the input size .

Model#ParamsGFLOPsPCKh

MobileNetV2 1×9.6M1.9785.4
MobileNetV3 1×8.7M1.8284.3
ShuffleNetV2 1×7.6M1.7082.8
Small HRNet-W161.3M0.7280.2
Lite-HRNet-181.1M0.2786.1
MobilePoseNet1.5M0.7487.3