Point Cloud Registration on ModelNet40 categories 1-20 (held-out models)
1.46MAE (Rotation)ReAgent (ours IL)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ReAgent (ours IL)Input features=3D coordinates, Training mode=Imitation Learning2021.03 | 1.46 | 0.011 | 2.82 | 0.023 | 94.5 | 0.75 | 21 | |
| ReAgent (ours IL+RL)Input features=3D coordinates, Training mode=Imitation Learning + Reinforcement Learning2021.03 | 1.47 | 0.011 | 2.87 | 0.023 | 94.5 | 0.75 | 21 | |
| PointNetLKInput features=3D coordinates2021.03 | 1.64 | 0.012 | 3.33 | 0.026 | 93 | 1.03 | 45 | |
| FGRInput features=3D coordinates and normals2021.03 | 2.52 | 0.016 | 4.37 | 0.034 | 92.1 | 1.59 | 68 | |
| DCP-v2Input features=3D coordinates2021.03 | 3.48 | 0.025 | 7.01 | 0.052 | 85.8 | 2.52 | 23 | |
| ICPInput features=3D coordinates2021.03 | 3.59 | 0.028 | 7.81 | 0.063 | 90.6 | 3.49 | 9 |