Table 5 Ablations of different modules and their contribution to the AP and the FPS on the KITTI dataset
From: Out-of-distribution- and location-aware PointNets for real-time 3D road user detection without a GPU
PVLE (cls) | PVLE (box) | IDP (cls) | IDP (box) | Cars AP | Pedestrians AP | Cyclists AP | Speed FPS | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Easy | Moderate | Hard | Easy | Moderate | Hard | Easy | Moderate | Hard | GPU | CPU | ||||
– | – | – | – | 44.44 | 39.27 | 36.81 | 34.37 | 31.32 | 29.31 | 39.43 | 31.31 | 30.10 | 64.5 | 6.5 |
– | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 48.12 | 49.87 | 47.43 | 43.83 | 39.96 | 34.86 | 48.68 | 39.84 | 39.07 | 76.1 | 15.3 |
\(\checkmark\) | – | \(\checkmark\) | \(\checkmark\) | 62.83 | 55.91 | 51.68 | 47.28 | 42.44 | 39.20 | 54.93 | 43.51 | 43.06 | 76.1 | 15.3 |
\(\checkmark\) | \(\checkmark\) | – | \(\checkmark\) | 58.25 | 42.43 | 38.55 | 36.42 | 30.36 | 26.73 | 45.20 | 37.45 | 39.98 | 64.5 | 6.4 |
\(\checkmark\) | \(\checkmark\) | \(\checkmark\) | – | 61.43 | 52.34 | 49.33 | 43.31 | 38.32 | 32.42 | 58.32 | 44.32 | 42.43 | 76.1 | 15.2 |
\(\checkmark\) | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 64.34 | 56.25 | 52.01 | 47.41 | 41.12 | 37.61 | 62.85 | 47.12 | 47.26 | 76.1 | 15.2 |