Visual Place Recognition on SPED
98.9Recall@1SAGE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAGEDim=8448, Two-stage processing=false2025.09 | 98.9 | 99.7 | 100 | — | |
| SAGE2025.09 | 98.9 | 99.7 | 100 | — | |
| SAGEDim=4096, Two-stage processing=false2025.09 | 97.7 | 99.8 | 100 | — | |
| SAGEDim=2048, Two-stage processing=false2025.09 | 95.6 | 99.2 | 99.7 | — | |
| EMVPDim=8448, Two-stage processing=false2025.09 | 94.6 | 97.5 | 98.4 | — | |
| DC-VLAQ2026.01 | 93.9 | 97.7 | 98.2 | — | |
| SuperVLADDim=3072, Two-stage processing=false2025.09 | 93.2 | 97 | 98 | — | |
| EffoVPR2024.05 | 93.1 | 97.9 | 98.4 | — | |
| EffoVPRTwo-stage reranking=true2025.09 | 93.1 | 97.9 | 98.4 | — | |
| FoLDim=N/A, Two-stage processing=true2025.09 | 92.6 | 96.5 | 97.4 | — | |
| BoQBackbones=DINOv2-B, Cd=122882025.07 | 92.5 | — | — | — | |
| BoQVenue & Year=CVPR'242026.01 | 92.5 | 95.9 | 96.7 | — | |
| BoQDim=12288, Two-stage processing=false2025.09 | 92.5 | 95.9 | 96.7 | — | |
| EDTformerDim=4096, Two-stage processing=false2025.09 | 92.4 | 95.9 | 96.9 | — | |
| EDTFormer2025.09 | 92.4 | 95.9 | 96.9 | — | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412023.11 | 92.1 | 96.2 | — | — | |
| SALAD2024.05 | 92.1 | 96.2 | 96.5 | — | |
| SALAD2025.04 | 92.1 | — | — | — | |
| FoLStage=global2025.04 | 92.1 | — | — | — | |
| SALADVenue & Year=CVPR'242026.01 | 92.1 | 96.2 | — | — | |
| FoL-globalVenue & Year=AAAI'252026.01 | 92.1 | — | — | — | |
| SALADDim=8448, Two-stage processing=false2025.09 | 92.1 | 96.2 | 96.5 | — | |
| FoLDim=8448, Two-stage processing=false2025.09 | 92.1 | 96.5 | 98 | — | |
| QAABackbones=DINOv2-B, Cd=81922025.07 | 91.8 | — | — | — | |
| FoLStage=re-ranking2025.04 | 91.8 | — | — | — | |
| VLAD-BuFF2025.09 | 91.4 | 95.9 | 96.9 | — | |
| CricaVPR2024.05 | 91.3 | 95.2 | 96.2 | — | |
| CricaVPR2025.04 | 91.3 | — | — | — | |
| CricaVPRDim=4096, Two-stage processing=false2025.09 | 91.3 | 95.2 | 96.2 | — | |
| QAABackbones=DINOv2-B, Cd=40962025.07 | 91.1 | — | — | — | |
| QAABackbones=DINOv2-B, Cd=20482025.07 | 91.1 | — | — | — | |
| QAABackbones=DINOv2-B, Cd=10242025.07 | 90.8 | — | — | — | |
| SuperVLADBackbone=DINOv2-B, ρ=/2026.05 | 90.6 | 95.4 | — | 7.38 | |
| WeiADBackbone=DINOv2-B, ρ=1.002026.05 | 90.6 | 95.4 | — | 2.41 | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.952026.05 | 90.4 | 95.3 | — | 2.23 | |
| SALADBackbone=DINOv2-B, ρ=/2026.05 | 90.3 | 95.4 | — | 2.41 | |
| UniPR-3DDesc. size=17152, Latency (ms)=8.232025.12 | 89.6 | 94.5 | — | — | |
| DINOv2 SALADDesc. size=2048 + 64, Latency (ms)=2.352023.11 | 89.5 | 94.9 | — | — | |
| SelaVPRVenue & Year=ICLR'242026.01 | 89.5 | — | — | — | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412025.12 | 89.5 | 94.4 | — | — | |
| SALAD-CMDim=8448, Two-stage processing=false2025.09 | 89.5 | 94.9 | 96.1 | — | |
| SALAD-CMVenue & Year=ECCV'242026.01 | 89.3 | — | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.702026.05 | 89.2 | 94.4 | — | 1.68 | |
| SALAD CMBackbones=DINOv2-B, Cd=84482025.07 | 89.1 | — | — | — | |
| SelaVPR2024.05 | 88.6 | 95.1 | 97.2 | — | |
| SelaVPR2025.04 | 88.6 | — | — | — | |
| SelaVPRDim=N/A, Two-stage processing=true2025.09 | 88.6 | 95.1 | 97.2 | — | |
| DINOv2 SALADDesc. size=512 + 32, Latency (ms)=2.332023.11 | 88.5 | 94.7 | — | — | |
| BoQ (Ours)Dim.=16384, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 86.5 | 93.4 | 95.7 | — | |
| BoQ2024.05 | 86.2 | 94.4 | 96.1 | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.502026.05 | 85.5 | 92.5 | — | 1.3 | |
| BoQ (Ours)Dim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 85.4 | 93.1 | 95.4 | — | |
| MixVPRBackbones=ResNet-50, Cd=40962025.07 | 85.2 | — | — | — | |
| MixVPRdim=4096, backbone=ResNet-502023.03 | 85.2 | 92.1 | 94.6 | — | |
| MixVPRDesc. size=4096, Latency (ms)=1.372023.11 | 85.2 | 92.1 | — | — | |
| MixVPRDim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 85.2 | 92.1 | 94.6 | — | |
| MixVPRVenue & Year=WACV'232026.01 | 85.2 | 92.1 | 94.6 | — | |
| MixVPRDesc. size=4096, Latency (ms)=1.372025.12 | 85.2 | 92.1 | — | — | |
| MixVPRBackbone=ResNet-50, ρ=/2026.05 | 85.2 | 92.1 | — | 1.37 | |
| MixVPRdim=2048, backbone=ResNet-502023.03 | 84.7 | 92.1 | 94.4 | — | |
| MixVPRDim=4096, Two-stage processing=false2025.09 | 84.7 | 92.3 | 94.4 | — | |
| SelaVPRDim=1024, Two-stage processing=false2025.09 | 83.5 | 92.6 | 94.6 | — | |
| EigenPlaceBackbones=ResNet-50, Cd=20482025.07 | 82.4 | — | — | — | |
| EigenPlacesDim.=2048, Backbone=ResNet-50, Pre-trained Weights=Authors' Weights2024.05 | 82.4 | 91.4 | 94.7 | — | |
| EigenPlacesVenue & Year=ICCV'232026.01 | 82.4 | 91.4 | 94.7 | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.402026.05 | 82.4 | 90.9 | — | 1.09 | |
| Conv-AP4x4Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=4x42022.10 | 81.2 | 90.3 | 93.9 | — | |
| Conv-AP3x3Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=3x32022.10 | 80.9 | 90.3 | 93.4 | — | |
| SFRSDim=4096, Two-stage processing=false2025.09 | 80.2 | 92.6 | 95.4 | — | |
| Conv-APBackbones=ResNet-50, Cd=40962025.07 | 80.1 | — | — | — | |
| Conv-AP2x2Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=2x22022.10 | 80.1 | 90.3 | 93.6 | — | |
| CosPlaceDesc. size=2048, Latency (ms)=2.592023.11 | 80.1 | 89.6 | — | — | |
| Conv-APDim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 80.1 | 90.3 | 93.6 | — | |
| CosPlaceDesc. size=2048, Latency (ms)=2.592025.12 | 80.1 | 89.6 | — | — | |
| CosPlaceBackbone=VGG-16, ρ=/2026.05 | 80.1 | 89.6 | — | 2.59 | |
| Conv-APDesc. size=8192, Latency (ms)=1.222023.11 | 79.2 | 88.6 | — | — | |
| Conv-APDesc. size=8192, Latency (ms)=1.222025.12 | 79.2 | 88.6 | — | — | |
| Conv-APBackbone=ResNet-50, ρ=/2026.05 | 79.2 | 88.6 | — | 1.22 | |
| NetVLADTraining Dataset=GSV-CITIES2022.10 | 78.7 | 88.3 | 91.4 | — | |
| NetVLADdim=32768, backbone=ResNet-502023.03 | 78.7 | 88.3 | 91.4 | — | |
| NetVLADDesc. size=32768, Latency (ms)=1.412023.11 | 78.7 | 88.3 | — | — | |
| NetVLADDim.=32768, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 78.7 | 88.3 | 91.4 | — | |
| NetVLADVenue & Year=CVPR'162026.01 | 78.7 | 88.3 | 91.4 | — | |
| NetVLADDesc. size=32768, Latency (ms)=1.412025.12 | 78.7 | 88.3 | — | — | |
| NetVLADBackbone=VGG-16, ρ=/2026.05 | 78.7 | 88.3 | — | 1.41 | |
| Gated NetVLADTraining Dataset=GSV-CITIES2022.10 | 75.6 | 87.1 | 90.8 | — | |
| Gated NetVLADdim=32768, backbone=ResNet-502023.03 | 75.6 | 87.1 | 90.8 | — | |
| Gated NetVLADDim.=32768, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 75.6 | 87.1 | 90.8 | — | |
| CosPlaceDim=512, Two-stage processing=false2025.09 | 75.5 | 87 | 89.6 | — | |
| CosPlaceBackbones=ResNet-50, Cd=20482025.07 | 75.3 | — | — | — | |
| CosPlaceTraining Dataset=GSV-CITIES2022.10 | 75.3 | 85.9 | 88.6 | — | |
| CosPlacedim=2048, backbone=ResNet-502023.03 | 75.3 | 85.9 | 88.6 | — | |
| CosPlaceDim.=2048, Backbone=ResNet-50, Pre-trained Weights=Authors' Weights2024.05 | 75.3 | 85.9 | 88.6 | — | |
| CosPlaceVenue & Year=CVPR'222026.01 | 75.3 | 85.9 | 88.6 | — | |
| Conv-AP1x1Training Dataset=GSV-CITIES, Adaptive Average Pooling Size (s x s)=1x12022.10 | 75 | 86.8 | 90.3 | — | |
| SPE-VLADTraining Dataset=GSV-CITIES2022.10 | 73.1 | 85.5 | 88.7 | — | |
| SPE-NetVLADdim=163840, backbone=ResNet-502023.03 | 73.1 | 85.5 | 88.7 | — | |
| SPE-NetVLADDim.=163840, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 73.1 | 85.5 | 88.7 | — | |
| NetVLADdim=32768, training_dataset=Pitts30k-train2023.03 | 71 | 87.1 | 90.4 | — | |
| EigenPlaces2024.05 | 70.2 | 83.5 | 87.5 | — |