Visual Place Recognition on Pitts 250k (test)
98.4Recall@1SAGE
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SAGEDim=84482025.09 | 98.4 | 99.4 | 99.7 | — | — | |
| SAGE2025.09 | 98.4 | 99.4 | 99.7 | — | — | |
| SAGEDim=40962025.09 | 98.2 | 99.4 | 99.5 | — | — | |
| SAGEDim=20482025.09 | 97.7 | 99.1 | 99.3 | — | — | |
| SuperVLADDim=30722025.09 | 97.3 | 99.4 | 99.7 | — | — | |
| EMVPDim=84482025.09 | 96.5 | 99.1 | 99.5 | — | — | |
| FoLDim=84482025.09 | 96.5 | 99.1 | 99.5 | — | — | |
| EDTFormer2025.09 | 95.9 | 98.8 | 99.3 | — | — | |
| SelaVPR2024.02 | 95.7 | 98.8 | 99.2 | — | — | |
| VLAD-BuFF2025.09 | 95.5 | 98.5 | 99.2 | — | — | |
| SALAD-CMDim=84482025.09 | 95.2 | 98.8 | 99.3 | — | — | |
| SuperVLADBackbone=DINOv2-B, ρ=/2026.05 | 95.2 | 98.6 | — | — | 7.38 | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412023.11 | 95.1 | 98.5 | — | — | — | |
| WeiADBackbone=DINOv2-B, ρ=1.002026.05 | 95.1 | 98.8 | — | — | 2.41 | |
| BoQ (Ours)Dim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 95 | 98.4 | 99.1 | — | — | |
| BoQ (Ours)Dim.=16384, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 95 | 98.5 | 99.1 | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.952026.05 | 95 | 98.6 | — | — | 2.23 | |
| UniPR-3DDesc. size=17152, Latency (ms)=8.232025.12 | 94.9 | 98.1 | — | — | — | |
| SALADBackbone=DINOv2-B, ρ=/2026.05 | 94.9 | 98.5 | — | — | 2.41 | |
| DINOv2 SALADDesc. size=2048 + 64, Latency (ms)=2.352023.11 | 94.8 | 98.3 | — | — | — | |
| MixVPRdim=4096, backbone=ResNet-502023.03 | 94.6 | 98.3 | 99 | — | — | |
| MixVPRDesc. size=4096, Latency (ms)=1.372023.11 | 94.6 | 98.3 | — | — | — | |
| MixVPRDesc. size=4096, Latency (ms)=1.372025.12 | 94.6 | 98.3 | — | — | — | |
| MixVPRBackbone=ResNet-50, ρ=/2026.05 | 94.6 | 98.3 | — | — | 1.37 | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412025.12 | 94.5 | 98.7 | — | — | — | |
| MixVPR2024.02 | 94.3 | 98.2 | 98.9 | — | — | |
| MixVPRDim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 94.2 | 98.2 | 98.9 | — | — | |
| MixVPRdim=2048, backbone=ResNet-502023.03 | 94.1 | 98.2 | 98.8 | — | — | |
| EigenPlacesDesc. size=2048, Latency (ms)=2.652023.11 | 94.1 | 98 | — | — | — | |
| EigenPlacesDim.=2048, Backbone=ResNet-50, Pre-trained Weights=Authors' Weights2024.05 | 94.1 | 97.9 | 98.7 | — | — | |
| EigenPlacesDesc. size=2048, Latency (ms)=2.652025.12 | 94.1 | 98 | — | — | — | |
| EigenPlacesBackbone=ResNet-50, ρ=/2026.05 | 94.1 | 98 | — | — | 2.65 | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.702026.05 | 94 | 98.3 | — | — | 1.68 | |
| DINOv2 SALADDesc. size=512 + 32, Latency (ms)=2.332023.11 | 93 | 97.4 | — | — | — | |
| Conv-APDesc. size=8192, Latency (ms)=1.222023.11 | 92.9 | 97.7 | — | — | — | |
| Conv-APDesc. size=8192, Latency (ms)=1.222025.12 | 92.9 | 97.7 | — | — | — | |
| Conv-APBackbone=ResNet-50, ρ=/2026.05 | 92.9 | 97.7 | — | — | 1.22 | |
| Conv-AP2x2Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=2x22022.10 | 92.4 | 97.4 | 98.4 | — | — | |
| Conv-APDim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 92.4 | 97.4 | 98.4 | — | — | |
| CosPlaceDim.=2048, Backbone=ResNet-50, Pre-trained Weights=Authors' Weights2024.05 | 92.3 | 97.4 | 98.4 | — | — | |
| Conv-AP3x3Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=3x32022.10 | 92.2 | 97.6 | 98.6 | — | — | |
| Conv-AP4x4Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=4x42022.10 | 92.2 | 97.4 | 98.3 | — | — | |
| CosPlaceDesc. size=2048, Latency (ms)=2.592023.11 | 92.1 | 97.5 | — | — | — | |
| CosPlaceDesc. size=2048, Latency (ms)=2.592025.12 | 92.1 | 97.5 | — | — | — | |
| CosPlaceBackbone=VGG-16, ρ=/2026.05 | 92.1 | 97.5 | — | — | 2.59 | |
| CosPlaceTraining Dataset=GSV-CITIES2022.10 | 91.5 | 96.9 | 97.9 | — | — | |
| CosPlacedim=2048, backbone=ResNet-502023.03 | 91.5 | 96.9 | 97.9 | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.502026.05 | 91.5 | 97.4 | — | — | 1.3 | |
| NetVLADTraining Dataset=GSV-CITIES2022.10 | 90.5 | 96.2 | 97.4 | — | — | |
| Conv-AP1x1Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=1x12022.10 | 90.5 | 96.2 | 97.5 | — | — | |
| NetVLADdim=32768, backbone=ResNet-50, results_source=re-trained2023.03 | 90.5 | 96.2 | 97.4 | — | — | |
| NetVLADDesc. size=32768, Latency (ms)=1.412023.11 | 90.5 | 96.2 | — | — | — | |
| NetVLADDim.=32768, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 90.5 | 96.2 | 97.4 | — | — | |
| NetVLADDesc. size=32768, Latency (ms)=1.412025.12 | 90.5 | 96.2 | — | — | — | |
| NetVLADBackbone=VGG-16, ρ=/2026.05 | 90.5 | 96.2 | — | — | 1.41 | |
| SPE-VLADTraining Dataset=GSV-CITIES2022.10 | 89.9 | 96.1 | 97.3 | — | — | |
| Gated NetVLADTraining Dataset=GSV-CITIES2022.10 | 89.7 | 95.9 | 97.1 | — | — | |
| Gated NetVLADdim=32768, backbone=ResNet-502023.03 | 89.7 | 95.9 | 97.1 | — | — | |
| Gated NetVLADDim.=32768, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 89.7 | 95.9 | 97.1 | — | — | |
| SPE-NetVLADdim=163840, backbone=ResNet-502023.03 | 89.2 | 95.3 | 97 | — | — | |
| SPE-NetVLADDim.=163840, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 89.2 | 95.3 | 97 | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.402026.05 | 88.2 | 96.2 | — | — | 1.09 | |
| GeMDesc. size=1024, Latency (ms)=1.142023.11 | 87 | 94.4 | — | — | — | |
| GeMDesc. size=1024, Latency (ms)=1.142025.12 | 87 | 94.4 | — | — | — | |
| NetVLADdim=32768, training_dataset=Pitts30k-train, results_source=reported by authors2023.03 | 86 | 93.2 | 95.1 | — | — | |
| GeMTraining Dataset=GSV-CITIES2022.10 | 82.9 | 92.1 | 94.3 | — | — | |
| GeMdim=2048, backbone=ResNet-50, results_source=re-trained2023.03 | 82.9 | 92.1 | 94.3 | — | — | |
| GeMDim.=2048, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 82.9 | 92.1 | 94.3 | — | — | |
| AVGTraining Dataset=GSV-CITIES2022.10 | 78.3 | 89.8 | 92.6 | — | — | |
| AVGdim=2048, backbone=ResNet-50, results_source=re-trained2023.03 | 78.3 | 89.8 | 92.6 | — | — | |
| AVGDim.=2048, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 78.3 | 89.8 | 92.6 | — | — | |
| GeMdim=2048, training_dataset=Pitts30k-train, results_source=reported by authors2023.03 | 72.3 | 87.2 | 91.4 | — | — | |
| AVGdim=2048, training_dataset=Pitts30k-train, results_source=reported by authors2023.03 | 62.6 | 82.7 | 88.4 | — | — | |
| DELFComputation time (msecs)=15872024.03 | — | — | — | 97 | — | |
| GB (SIFT)Computation time (msecs)=27092024.03 | — | — | — | 92 | — | |
| L2-distComputation time (msecs)=0.052024.03 | — | — | — | 87 | — | |
| SIFTComputation time (msecs)=1292024.03 | — | — | — | 92 | — | |
| STUNComputation time (msecs)=0.12024.03 | — | — | — | 79 | — | |
| SUEComputation time (msecs)=1.082024.03 | — | — | — | 94 | — |