Relative Pose Estimation on ScanNet SuperGlue (test)
0.99VCRE AUCLoFTR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LoFTRTraining Signal=depth (D) + overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.99 | 91.3 | 0.13 | 1.81 | |
| ASpanFormerTraining Signal=depth (D) + overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.99 | 90.6 | 0.14 | 1.48 | |
| RoMaTraining Signal=depth (D) + overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.99 | 94.4 | 0.11 | 1.47 | |
| MicKey-OTraining Signal=overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.99 | 93.7 | 0.17 | 3.58 | |
| MicKeyTraining Signal=pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.99 | 92.8 | 0.17 | 3.64 | |
| SuperGlueTraining Signal=depth (D) + overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.98 | 90.6 | 0.15 | 2.06 | |
| RPR [3D-3D]Training Signal=overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.93 | 92.8 | 0.2 | 4.05 | |
| RPR [R + s-t]Training Signal=overlap score (O) + pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.85 | 85 | 0.25 | 4.12 | |
| RPR [3D-3D]Training Signal=pose (P), Scale Recovery=PlaneRCNN2024.04 | 0.78 | 78.3 | 0.55 | 5.12 |