Visual Localization on Cambridge Landmarks Church
0.04Median Translation Error (m)MASt3R + Kapture
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MASt3R + KaptureTraining Category=Seen, Latency=9.0s2025.10 | 0.04 | 0.1 | — | |
| hlocMethod Category=SOTA, Training Protocol=TL2024.04 | 0.07 | 0.21 | — | |
| hLoc (SP+SG)Training Category=Seen, Latency=∼1.2s, Mapping=~35min2025.10 | 0.07 | 0.2 | — | |
| ASMethod Category=SOTA2024.04 | 0.08 | 0.25 | — | |
| AS (SIFT)Training Category=Seen, Latency=0.4s, Mapping=~35min2025.10 | 0.08 | 0.3 | — | |
| HACNetMethod Category=SOTA, Training Protocol=TS2024.04 | 0.09 | 0.3 | — | |
| GLACETraining Category=Seen, Latency=0.1s, Mapping=25min2025.10 | 0.09 | 0.3 | — | |
| E5+1 (ALKD-LG)Training Category=Unseen, Latency=1.1s, Mapping=~30s2025.10 | 0.09 | 0.3 | — | |
| PixLocMethod Category=SOTA, Training Protocol=TL2024.04 | 0.1 | 0.34 | — | |
| DSAC*Method Category=SOTA, Training Protocol=TS2024.04 | 0.13 | 0.4 | — | |
| FastForwardTraining Category=Unseen, Latency=0.5s, Mapping=~30s2025.10 | 0.14 | 0.5 | — | |
| ACETraining Category=Seen, Latency=0.1s, Mapping=5min2025.10 | 0.2 | 0.6 | — | |
| HLocLocalization Approach=FM, Scene Compression=false2022.11 | 0.21 | 0.07 | 2,007 | |
| MCLocMethod Category=Pose Refiners2024.04 | 0.26 | 0.88 | — | |
| HSCNetLocalization Approach=E2E, Scene Compression=false2022.11 | 0.3 | 0.09 | 207 | |
| NeFeS (DFNet)Method Category=Pose Refiners, Training Protocol=TS2024.04 | 0.32 | 0.99 | — | |
| Reloc3rTraining Category=Unseen, Latency=0.6s, Mapping=~30s2025.10 | 0.33 | 0.6 | — | |
| DSMLocalization Approach=E2E, Scene Compression=false2022.11 | 0.34 | 0.11 | 174 | |
| SqueezerLocalization Approach=FM, Scene Compression=true2022.11 | 0.37 | 0.15 | 7.97 | |
| CROSSFIREMethod Category=Pose Refiners, Training Protocol=TS2024.04 | 0.39 | 1.4 | — | |
| DSAC++Localization Approach=E2E, Scene Compression=false2022.11 | 0.4 | 0.13 | 207 | |
| HybridLocalization Approach=FM, Scene Compression=true2022.11 | 0.49 | 0.5 | 1.34 | |
| NeuMapLocalization Approach=E2E, Scene Compression=true2022.11 | 0.53 | 0.17 | 0.4 | |
| LENSMethod Category=Pose Regressors, Training Protocol=TS2024.04 | 0.53 | 1.6 | — | |
| FQNMethod Category=Pose Refiners, Training Protocol=TS2024.04 | 0.58 | 2 | — | |
| Active searchLocalization Approach=FM, Scene Compression=false2022.11 | 0.62 | 0.22 | 359 | |
| QP+RootSIFTLocalization Approach=FM, Scene Compression=true2022.11 | 0.89 | 0.56 | 3.3 | |
| DFNetMethod Category=Pose Regressors, Training Protocol=TS2024.04 | 1.37 | 4.03 | — | |
| Relpose-GNNTraining Category=Unseen, Latency=N/A, Mapping=~30s2025.10 | 1.52 | 3.2 | — | |
| MS-TransformerMethod Category=Pose Regressors, Training Protocol=TS2024.04 | 1.62 | 3.99 | — | |
| DenseVLADMethod Category=Retrieval2024.04 | 2.3 | 8 | — | |
| ExReNetTraining Category=Unseen, Latency=∼0.4s, Mapping=~30s2025.10 | 2.3 | 3.7 | — | |
| Map-freeTraining Category=Unseen, Latency=∼0.2s, Mapping=~30s2025.10 | 2.91 | 5.1 | — | |
| CosPlaceMethod Category=Retrieval2024.04 | 3.2 | 7.2 | — | |
| MASt3RTraining Category=Unseen, Latency=9.0s, Mapping=~30s2025.10 | 3.43 | 0.7 | — |