Class-agnostic 3D Instance Segmentation on Waymo Open Dataset (WOD) (val)
45.7AP70fully manual annotation
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| fully manual annotationannotation setting=supervised, #point clouds=158k, #3D bboxes=4268k2022.10 | 45.7 | 37.3 | 75.1 | 65.1 | 92.2 | |
| ClusterNetannotation setting=unsupervised, motion cues=ground truth scene flow, #point clouds=158k, #3D bboxes=02022.10 | 42 | 33.2 | 61.7 | 52.3 | 88.1 | |
| ClusterNetannotation setting=unsupervised, #point clouds=158k, #3D bboxes=02022.10 | 26.2 | 19.2 | 40 | 32.8 | 64.9 | |
| 10% manual annotationannotation setting=supervised, #point clouds=15k, #3D bboxes=397k2022.10 | 23.6 | 15.5 | 61.8 | 48.7 | 81.6 |