3D registration on DirLab landmarks 3,000 expert-annotated (test)
2.86Average Error (mm)D-RobOT (LDDMM)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| D-RobOT (LDDMM)Training regime=Supervised, Input modality=Point clouds, Regularization=LDDMM2021.11 | 2.86 | 1.25 | 2.23 | 3.11 | 1.92 | |
| D-RobOT (spline)Training regime=Supervised, Input modality=Point clouds, Regularization=Spline2021.11 | 2.95 | 1.3 | 2.5 | 3.19 | 1.87 | |
| D-RobOT (raw)Training regime=Supervised, Input modality=Point clouds, Variation=Raw2021.11 | 3.4 | 1.4 | 2.58 | 3.69 | 1.26 | |
| DispEmd+Training regime=Supervised, Input modality=Image keypoints2021.11 | 3.42 | — | — | — | — | |
| DGCNN-CPD+Training regime=Supervised, Input modality=Image keypoints2021.11 | 4.3 | — | — | — | — | |
| S-RobOT (LDDMM)Training regime=Supervised, Input modality=Point clouds, Regularization=LDDMM2021.11 | 5.48 | 2.86 | 4.44 | 7.14 | 42.3 | |
| S-RobOT (spline)Training regime=Supervised, Input modality=Point clouds, Regularization=Spline2021.11 | 5.72 | 3.19 | 5.04 | 7.35 | 2.77 | |
| CPD (non-rigid)Training regime=No training, Input modality=Point clouds, Point cloud density=20k, Variation=Non-rigid2021.11 | 9.3 | 5.95 | 8.6 | 11.83 | 332.6 | |
| RobOT (raw)Training regime=No training, Input modality=Point clouds, Variation=Raw2021.11 | 9.41 | 4.89 | 8.35 | 13.04 | 0.15 | |
| RobOT (affine)Training regime=No training, Input modality=Point clouds, Variation=Affine2021.11 | 10.45 | 6.01 | 9.83 | 13.97 | 0.18 | |
| ICP (affine)Training regime=No training, Input modality=Point clouds, Variation=Affine2021.11 | 15.05 | 9.6 | 14.06 | 20.01 | 0.52 | |
| Input dataTraining regime=No training, Input modality=Point clouds2021.11 | 23.3 | 13.18 | 22.22 | 31.65 | — |