Research Article
Pillar-Based 3D Object Detection from Point Cloud with Multiattention Mechanism
Table 1
Results on the KITTI test BEV detection benchmark.
| Model | Car | Pedestrian | Cyclist | Easy (%) | Mod. (%) | Hard (%) | mAP (%) | Easy (%) | Mod. (%) | Hard (%) | mAP (%) | Easy (%) | Mod. (%) | Hard (%) | mAP (%) |
| PointPillars | 88.35 | 86.10 | 79.83 | 84.76 | 58.66 | 50.23 | 47.19 | 52.02 | 79.14 | 62.25 | 56.00 | 65.79 | PointPillars + point-wise attention | 89.63 | 86.72 | 83.72 | 86.69 | 58.62 | 53.29 | 48.69 | 53.53 | 83.69 | 63.04 | 61.62 | 69.45 | PointPillars + SOPA | 91.23 | 87.27 | 84.90 | 87.80 | 58.11 | 52.24 | 49.12 | 53.16 | 81.72 | 66.96 | 62.60 | 70.43 |
|
|