Table 1 Performance comparison results of 3D target detection models in the nuScenes test set
From: CenterTransFuser: radar point cloud and visual information fusion for 3D object detection
Methods | MDL | NDS (%) | mAP (%) | mATE (%) | mASE (%) | mAOE (%) | mAVE (%) | mAAE (%) |
|---|---|---|---|---|---|---|---|---|
PointPillars [21] | L | 0.453 | 0.305 | 0.517 | 0.290 | 0.500 | 0.316 | 0.319 |
SARPNET [35] | L | 0.484 | 0.324 | 0.400 | 0.249 | 0.763 | 0.272 | 0.090 |
InfoFocus [36] | L | 0.395 | 0.375 | 0.363 | 0.265 | 1.132 | 1.000 | 0.395 |
MonoDIS [18] | C | – | 0.304 | 0.738 | 0.263 | 0.546 | 1.553 | 0.134 |
CenterNet [19] | C | 0.400 | 0.338 | 0.658 | 0.255 | 0.629 | 1.629 | 0.142 |
SPRCNN [4] | L + C | – | 0.361 | 0.751 | 0.231 | 0.571 | 1.672 | 0.112 |
CenterFusion [11] | R + C | 0.449 | 0.326 | 0.631 | 0.261 | 0.516 | 0.614 | 0.115 |
Our model | R + C | 0.471 | 0.347 | 0.628 | 0.252 | 0.523 | 0.527 | 0.135 |