TABLE 7.
Experiment result showing the average recall (%) at top 1% for each of the models. Both methods are only trained on Oxford (Maddern et al., 2017) and tested on other different data sets (Uy and Lee, 2018).
| Datasets | Oxford | U.S. | R.A. | B.D. |
| PointNetVLAD Uy and Lee, (2018) | 84.94% | 80.79% | 73.86% | 69.29% |
| EPN-NetVLAD (Ours) | 89.17% | 87.82% | 81.98% | 76.91% |
Bold values in tables show the best performance.