Table 5 Sensitivity analysis of BAP-MDSNA module parameters.
From: Learning spatio-temporal context for basketball action pose estimation with a multi-stream network
Experiment Condition | Head AP (%) | Shou AP (%) | Wri AP (%) | Overall MAP (%) |
|---|---|---|---|---|
Baseline (ResNet-50) | 82.1 | 85.3 | 76.1 | 80.2 |
+ VGG-16 | 81.3 | 84.7 | 75.6 | 79.5 |
+ ResNet-101 | 82.5 | 85.6 | 76.4 | 80.5 |
+ EfficientNet-B3 | 82.0 | 85.1 | 75.9 | 80.0 |
Higher resolution (\(512 \times 512\)) | 82.8 | 85.8 | 76.8 | 81.0 |
Lower resolution (\(128 \times 128\)) | 79.9 | 83.2 | 74.2 | 78.4 |
