Table 2.

Object keypoint similarity metrics (OKS) resulting from the different keypoint detection models stating the average precision with threshold values of 0.50 (AP₅₀) and 0.75 (AP₇₅) and averaged over thresholds from 0.5 to 0.95 (AP) as well as the average recall with the threshold values of 0.50 (AR₅₀) and 0.75 (AR₇₅). These were averaged over thresholds from 0.50 to 0.95 (AR). Best-performing values are printed in bold.

Architecture Type	AP0.50	AP0.75	AP	AR0.50	AR0.75	AR
Baseline–ResNet50	0.648	0.107	0.213	0.691	0.198	0.292
Baseline–ResNet101	0.640	0.107	0.228	0.687	0.200	0.288
Baseline–ResNet152	0.659	0.134	0.254	0.703	0.231	0.313
HRNet-W32	0.692	0.158	0.267	0.726	0.241	0.323
HRNet-W48	0.735	0.246	0.322	0.762	0.355	0.383