Table 5 Performance evaluation results. Semantic segmentation task (per point level) is evaluated using the \(F_1\) score. Instance segmentation task is evaluated using AP at varying IoU thresholds. The value of each metric is reported in %. Bold shows the overall performance averaged across classes.
From: Deep segmentation of 3+1D radar point cloud for real-time roadside traffic user detection
| Â | \(\varvec{F}_1\) | AP.30 | AP.50 | AP.75 | AP |
|---|---|---|---|---|---|
Background | 98.78 | – | – | – | – |
Person | 96.13 | 94.80 | 92.67 | 84.10 | 84.83 |
Bicycle | 96.65 | 94.28 | 92.44 | 84.73 | 85.16 |
Motorcycle | 89.18 | 95.73 | 91.03 | 85.27 | 82.83 |
Car | 94.85 | 94.33 | 93.00 | 77.73 | 76.90 |
Bus | 96.51 | 87.63 | 86.03 | 75.51 | 70.35 |
Average | 95.35 | 93.36 | 91.03 | 81.47 | 80.01 |
