Skip to main content
. 2023 Apr 4;23(7):3725. doi: 10.3390/s23073725

Table 4.

BAPose results and comparison with SOTA methods for the COCO dataset for test-dev. Best results are in bold.

Method Input
Size
AP AP50 AP75 APM APL AR
Single-Scale Testing
BAPose-W48 (ours) 640 70.3% 89.6% 77.5% 65.9% 77.1% 75.4%
HRNet-W48 [6] 640 70.0% 89.4% 77.3% 65.7% 76.9% 75.4%
HigherHRNet-W48 [14] 640 68.4% 88.2% 75.1% 64.4 74.2 -
BAPose-W32 (ours) 512 68.0% 88.0% 74.8% 62.4% 76.6% 72.9%
HRNet-W32 [6] 512 67.3% 87.9% 74.1% 61.5% 76.1% 72.4%
SPM [31] - 66.9% 88.5% 72.9% 62.6% 73.1% -
CenterNet-HG [42] 512 63.0% 86.8% 69.6% 58.9% 70.4% -
OpenPose [5] - 61.8% 84.9% 67.5% 57.1% 68.2% 66.5%
Multi-Scale Testing
BAPose-W48 (ours) 640 71.2% 89.4% 78.1% 67.4% 76.8% 76.8%
HRNet-W48 [6] 640 71.0% 89.2% 78.0% 67.1% 76.9% 76.7%
HigherHRNet-W48 [14] 640 70.5% 89.3% 77.2% 66.6% 75.8% -
Point-set Anchors [43] 640 68.7% 89.9% 76.3% 64.8% 75.3% 74.8%
BAPose-W32 (ours) 512 70.4% 89.3% 77.4% 66.0% 76.9% 75.6%
HRNet-W32 [6] 512 69.6% 89.0% 76.6% 65.2% 76.5% 75.1%
HGG [30] 512 67.6% 85.1% 73.7% 62.7% 74.6% 71.3%