Two-frame relative pose estimation on CMU Panoptic Studio (test)
18.21AUC @15°Our SfM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Our SfMcomponents=Virtual Correspondences (VCs) + classic correspondences2022.06 | 18.21 | 46.05 | 62.08 | |
| Deep regression2022.06 | 14.36 | 18.6 | 23.18 | |
| SuperGlueRefinement=Bundle Adjustment (BA)2022.06 | 10.68 | 16.57 | 18.92 | |
| SuperGluepose estimation algorithm=five-point algorithm2022.06 | 10.02 | 16.74 | 19.36 | |
| SuperPointRefinement=Bundle Adjustment (BA)2022.06 | 9.22 | 13.77 | 15.85 | |
| LoFTRRefinement=Bundle Adjustment (BA)2022.06 | 8.35 | 14.52 | 17.01 | |
| Deep optimization2022.06 | 7.88 | 27.17 | 42.42 | |
| SIFTRefinement=Bundle Adjustment (BA)2022.06 | 7.68 | 11.39 | 13.33 | |
| LoFTRpose estimation algorithm=five-point algorithm2022.06 | 5.12 | 10.47 | 13.07 |