Point cloud place recognition on U.S. reference maps in-house (test)
94.45Avg Recall @ 1%PN_VLAD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PN_VLADTraining Protocol=Refined (Oxford, U.S., R.A.)2019.04 | 94.45 | 86.06 | |
| PCANTraining Protocol=Refined (Oxford, U.S., R.A.)2019.04 | 94.07 | 83.69 | |
| PN_VLADArchitecture=PointNetVLAD, Training Datasets=Oxford, U.S., R.A.2018.04 | 90.1 | 86.07 | |
| PN_MAXArchitecture=PointNet + Max Pooling, Training Datasets=Oxford, U.S., R.A.2018.04 | 79.31 | 62.16 | |
| PointNetVLAD (PN_VLAD)Clusters (K)=64, Loss=lazy quadruplet loss, Output Dimension=2562018.04 | 72.63 | — | |
| PN_MAXLoss=lazy quadruplet loss, Output Dimension=2562018.04 | 64.64 | — | |
| PN_STDOutput Dimension=2562018.04 | 61.12 | — | |
| PN_STDArchitecture=PointNet + Standard Deviation, Training Datasets=Oxford, U.S., R.A.2018.04 | 56.95 | 45.67 |