3D human pose estimation model using location-maps for distorted and disconnected images by a wearable omnidirectional camera

Table 5 MPJPE performance comparison in different scales (1x, 2x, and 4x) of maps on the validation and test datasets-A

Model	Backbone	Out-scale	Head	Neck	Torso	Waist	Shoulders	Elbows	Wrists	Hands	All
Validation dataset-A
Ours (1x)	HRNet-W24	24 x 48	0.018	0.000	0.021	0.022	0.012	0.037	0.051	0.058	0.031
Ours (2x)	HRNet-W24	12 x 24	0.019	0.000	0.019	0.021	0.013	0.039	0.052	0.059	0.032
Ours (4x)	HRNet-W24	6 x 12	0.023	0.001	0.022	0.024	0.017	0.043	0.055	0.064	0.036
Test dataset-A
Ours (1x)	HRNet-W24	24 x 48	0.052	0.000	0.053	0.066	0.029	0.160	0.296	0.349	0.154
Ours (2x)	HRNet-W24	12 x 24	0.053	0.000	0.051	0.071	0.030	0.170	0.330	0.379	0.166
Ours (4x)	HRNet-W24	6 x 12	0.050	0.000	0.051	0.065	0.036	0.166	0.298	0.338	0.153