💡PCA features visualization for a pair of LiDAR and image in the validation set.
Top row: PCA features of DINO Bottom row: PCA features of projected voxel features
VXP v2 outperforms VXP in cross-modal place recognition
| Recall@1 | 2D-2D | 3D-3D | 2D-3D | 3D-2D |
|---|---|---|---|---|
| Cattaneo | 88.40 | 93,99 | 41,92 | 29,51 |
| LC2 | - - | - - | - - | - - |
| liploc | 59.5744 | 70.4453 | 32.6137 | 26.7000 |
| vxp | 92.0218 | 94.6916 | 44.6033 | 30.8090 |
| vxpv2 | 88.5861 | 96.2959 | 63.4741 | 58.2897 |
| Recall@1% | 2D-2D | 3D-3D | 2D-3D | 3D-2D |
|---|---|---|---|---|
| Cattaneo | 96.6 | 98.4 | 77.3 | 70.4 |
| LC2 | 84.1 | 83.0 | 81.2 | 73.8 |
| liploc | 90.2036 | 92.2567 | 77.7881 | 73.6100 |
| vxp | 98.7965 | 98.8433 | 84.3917 | 76.9310 |
| vxpv2 | 98.2309 | 99.4591 | 95.1170 | 93.9706 |
VXP outperforms VXP v2 in both uni and cross-modal place recognition
| Recall@1 | 2D-2D | 3D-3D | 2D-3D | 3D-2D |
|---|---|---|---|---|
| LC2 | 69.2 | 58.1 | 60.9 | 51.8 |
| cattaneo | 93.38925 | 90.97315 | 87.58360 | 78.5841 |
| liploc | 61.10215 | 78.76445 | 73.69395 | 54.9198 |
| vxp | 96.67990 | 96.98190 | 96.78070 | 94.5331 |
| vxpv2 | 93.35920 | 89.20055 | 84.97145 | 84.5016 |
| Recall@1% | 2D-2D | 3D-3D | 2D-3D | 3D-2D |
|---|---|---|---|---|
| LC2 | 96.9 | 96.1 | 96.0 | 94.6 |
| cattaneo | 99.8322 | 99.89935 | 99.63085 | 98.62375 |
| liploc | 93.9596 | 97.38295 | 98.42260 | 93.01880 |
| vxp | 99.8994 | 99.73170 | 99.63110 | 99.79880 |
| vxpv2 | 99.3293 | 98.62510 | 98.18890 | 98.45735 |