Visual Place Recognition on MSLS (val)
98.1Recall@1QAA
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| QAABackbone=DINOv2-B, Cd=40962025.07 | 98.1 | — | — | — | — | — | — | — | — | — | — | |
| QAABackbone=DINOv2-B, Cd=20482025.07 | 97.8 | — | — | — | — | — | — | — | — | — | — | |
| QAABackbone=DINOv2-B, Cd=10242025.07 | 97.7 | — | — | — | — | — | — | — | — | — | — | |
| QAABackbone=DINOv2-B, Cd=81922025.07 | 97.6 | — | — | — | — | — | — | — | — | — | — | |
| Pair-VPR-pEnc. Size=ViT-G2024.10 | 95.4 | 97.3 | 97.7 | — | — | — | — | — | — | — | — | |
| Pair-VPRAttack=None2026.06 | 95.4 | 97.3 | — | — | — | — | — | — | — | — | — | |
| SAGEDim=8448, Two-stage processing=false2025.09 | 94.5 | 97.4 | 97.8 | — | — | — | — | — | — | — | — | |
| SAGE2025.09 | 94.5 | 97.4 | 97.8 | — | — | — | — | — | — | — | — | |
| SALAD CMBackbone=DINOv2-B, Cd=84482025.07 | 94.2 | — | — | — | — | — | — | — | — | — | — | |
| SALAD-CMVenue & Year=ECCV’242026.01 | 94.2 | 97.2 | 97.4 | — | — | — | — | — | — | — | — | |
| DC-VLAQVenue & Year=-2026.01 | 94.2 | 97.3 | 97.6 | — | — | — | — | — | — | — | — | |
| SALAD-CMDim=8448, Two-stage processing=false2025.09 | 94.2 | 97.2 | 97.4 | — | — | — | — | — | — | — | — | |
| SALAD-CMRes=322, Two-stage=false2026.04 | 94.2 | 97.2 | 97.4 | — | — | — | — | — | — | — | — | |
| FoL++Matching Time (s)=0.047, Inference Resolution=322 × 322, Device=NVIDIA RTX A6000, Re-ranking Protocol=Two-Stage VPR2026.04 | 94.1 | 97.8 | 98.1 | — | — | — | — | — | — | — | — | |
| FoL++Res=322, Two-stage=true2026.04 | 94.1 | 97.8 | 98.1 | — | — | — | — | — | — | — | — | |
| EMVPDim=8448, Two-stage processing=false2025.09 | 93.9 | 97.3 | 97.6 | — | — | — | — | — | — | — | — | |
| BoQBackbone=DINOv2-B, Cd=122882025.07 | 93.8 | — | — | — | — | — | — | — | — | — | — | |
| BoQVenue & Year=CVPR’242026.01 | 93.8 | 96.8 | 97 | — | — | — | — | — | — | — | — | |
| BoQDim=12288, Two-stage processing=false2025.09 | 93.8 | 96.8 | 97 | — | — | — | — | — | — | — | — | |
| BoQRes=322, Two-stage=false2026.04 | 93.8 | 96.8 | 97 | — | — | — | — | — | — | — | — | |
| Pair-VPR-sEnc. Size=ViT-B2024.10 | 93.7 | 97.2 | 97.3 | — | — | — | — | — | — | — | — | |
| SAGEDim=4096, Two-stage processing=false2025.09 | 93.7 | 97.3 | 97.8 | — | — | — | — | — | — | — | — | |
| FoLDim=N/A, Two-stage processing=true2025.09 | 93.5 | 96.9 | 97.6 | — | — | — | — | — | — | — | — | |
| MegaLocRes=322, Two-stage=false2026.04 | 93.5 | 97 | 97.8 | — | — | — | — | — | — | — | — | |
| FoLAttack=None2026.06 | 93.5 | 96.9 | — | — | — | — | — | — | — | — | — | |
| SAGERes=322, Two-stage=false2026.04 | 93.4 | 97.3 | 97.6 | — | — | — | — | — | — | — | — | |
| FoL++ (global)Res=322, Two-stage=false2026.04 | 93.2 | 98 | 98.2 | — | — | — | — | — | — | — | — | |
| FoL-globalVenue & Year=AAAI’252026.01 | 93.1 | 96.9 | 97.4 | — | — | — | — | — | — | — | — | |
| FoLDim=8448, Two-stage processing=false2025.09 | 93.1 | 96.9 | 97.4 | — | — | — | — | — | — | — | — | |
| ImAgeRes=322, Two-stage=false2026.04 | 93 | 97 | 97.2 | — | — | — | — | — | — | — | — | |
| A2GCAttack=None2026.06 | 93 | 97.3 | — | — | — | — | — | — | — | — | — | |
| EffoVPR-Rtwo-stage=true2024.05 | 92.8 | 97.2 | 97.4 | — | — | — | — | — | — | — | — | |
| EffoVPRTwo-stage reranking=true2025.09 | 92.8 | 97.2 | 97.4 | — | — | — | — | — | — | — | — | |
| FoL (global)Res=322, Two-stage=false2026.04 | 92.8 | 96.9 | 97.2 | — | — | — | — | — | — | — | — | |
| SAGEDim=2048, Two-stage processing=false2025.09 | 92.6 | 96.9 | 97.7 | — | — | — | — | — | — | — | — | |
| AsymVPRQuery Network=EfficientViT-B2, Gallery Model Training Configuration=BoQ (DINOv2-B), Retrieval Protocol=Asymmetric2025.12 | 92.3 | 95.9 | — | — | — | — | — | — | — | — | — | |
| DINOv2 SALADDesc. size (Global)=8192 + 256, Desc. size (Local)=0.0, Memory (GB)=0.63, Latency (ms) Retrieval=2.41, Latency (ms) Reranking=0.02023.11 | 92.2 | 96.4 | 97 | — | — | — | — | — | — | — | — | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412023.11 | 92.2 | 96.4 | — | — | — | — | — | — | — | — | — | |
| SALADDimension=84482024.05 | 92.2 | — | — | — | — | — | — | — | — | — | — | |
| SALADtwo-stage=false2024.05 | 92.2 | 96.2 | 97 | — | — | — | — | — | — | — | — | |
| SALADEnc. Size=ViT-B2024.10 | 92.2 | 96.4 | 97 | — | — | — | — | — | — | — | — | |
| SALADVenue & Year=CVPR’242026.01 | 92.2 | 96.4 | 97 | — | — | — | — | — | — | — | — | |
| SALADDim=8448, Two-stage processing=false2025.09 | 92.2 | 96.4 | 97 | — | — | — | — | — | — | — | — | |
| SuperVLADDim=3072, Two-stage processing=false2025.09 | 92.2 | 96.6 | 97.4 | — | — | — | — | — | — | — | — | |
| SALADRes=322, Two-stage=false2026.04 | 92.2 | 96.4 | 97 | — | — | — | — | — | — | — | — | |
| SuperVLADRes=322, Two-stage=false2026.04 | 92.2 | 96.6 | 97.4 | — | — | — | — | — | — | — | — | |
| SALADAttack=None2026.06 | 92.2 | 96.2 | — | — | — | — | — | — | — | — | — | |
| EDTFormerVenue & Year=TCSVT’252026.01 | 92 | 96.6 | 97.2 | — | — | — | — | — | — | — | — | |
| SALADQuery Network=DINOv2-B, Gallery Model Training Configuration=SALAD (DINOv2-B), Retrieval Protocol=Symmetric2025.12 | 92 | 96.4 | — | — | — | — | — | — | — | — | — | |
| EDTformerDim=4096, Two-stage processing=false2025.09 | 92 | 96.6 | 97.2 | — | — | — | — | — | — | — | — | |
| EDTFormer2025.09 | 92 | 96.6 | 97.2 | — | — | — | — | — | — | — | — | |
| EDTFormerRes=322, Two-stage=false2026.04 | 92 | 96.6 | 97.2 | — | — | — | — | — | — | — | — | |
| VLAD-BuFF2025.09 | 91.8 | 96 | 96.2 | — | — | — | — | — | — | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.952026.05 | 91.8 | 96.3 | — | — | — | — | — | — | — | — | 2.23 | |
| WeiADBackbone=DINOv2-B, ρ=1.002026.05 | 91.7 | 96.2 | — | — | — | — | — | — | — | — | 2.41 | |
| SuperVLADBackbone=DINOv2-B, ρ=/2026.05 | 91.5 | 95.9 | — | — | — | — | — | — | — | — | 7.38 | |
| BoQExtrac. (ms)=7, Rerank. (ms)=02024.05 | 91.4 | 94.5 | — | — | — | — | — | — | — | — | — | |
| UniPR-3DDesc. size=17152, Latency (ms)=8.232025.12 | 91.4 | 96 | — | — | — | — | — | — | — | — | — | |
| SALADBackbone=DINOv2-B, ρ=/2026.05 | 91.4 | 96.2 | — | — | — | — | — | — | — | — | 2.41 | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.702026.05 | 91.3 | 96.1 | — | — | — | — | — | — | — | — | 1.68 | |
| BoQ (Ours)Dim.=16384, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 91.2 | 95.3 | 96.1 | — | — | — | — | — | — | — | — | |
| BoQEnc. Size=ResNet-502024.10 | 91.2 | 95.3 | 96.1 | — | — | — | — | — | — | — | — | |
| DINOv2 SALADDesc. size=8192 + 256, Latency (ms)=2.412025.12 | 91.2 | 95.3 | — | — | — | — | — | — | — | — | — | |
| AsymVPRQuery Network=EfficientViT-B2, Gallery Model Training Configuration=SALAD (DINOv2-B), Retrieval Protocol=Asymmetric2025.12 | 91.2 | 95.8 | — | — | — | — | — | — | — | — | — | |
| BoQ (Ours)Dim.=4096, Backbone=ResNet-50, Training Dataset=GSV-Cities2024.05 | 91.1 | 94.8 | 95.7 | — | — | — | — | — | — | — | — | |
| AsymVPRQuery Network=MobileViTv2, Gallery Model Training Configuration=BoQ (DINOv2-B), Retrieval Protocol=Asymmetric2025.12 | 91.1 | 96.1 | — | — | — | — | — | — | — | — | — | |
| EffoVPR-GDimension=10242024.05 | 90.9 | — | — | — | — | — | — | — | — | — | — | |
| SelaVPRStage=Two-stage2024.02 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRtwo-stage=true2024.05 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPREnc. Size=ViT-L2024.10 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRVenue & Year=ICLR’242026.01 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRDim=N/A, Two-stage processing=true2025.09 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRMatching Time (s)=0.085, Inference Resolution=322 × 322, Device=NVIDIA RTX A6000, Re-ranking Protocol=Two-Stage VPR2026.04 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRRes=322, Two-stage=true2026.04 | 90.8 | 96.4 | 97.2 | — | — | — | — | — | — | — | — | |
| SelaVPRAttack=None2026.06 | 90.8 | 96.4 | — | — | — | — | — | — | — | — | — | |
| BoQDimension=163842024.05 | 90.7 | — | — | — | — | — | — | — | — | — | — | |
| BoQtwo-stage=false2024.05 | 90.7 | 94.7 | 95.8 | — | — | — | — | — | — | — | — | |
| DINOv2 SALADDesc. size=2048 + 64, Latency (ms)=2.352023.11 | 90.5 | 95.4 | — | — | — | — | — | — | — | — | — | |
| WeiAD + WeiToPBackbone=DINOv2-B, ρ=0.502026.05 | 90.4 | 95.9 | — | — | — | — | — | — | — | — | 1.3 | |
| EffoVPR-GDimension=2562024.05 | 90.1 | — | — | — | — | — | — | — | — | — | — | |
| FoLMatching Time (s)=0.078, Inference Resolution=322 × 322, Device=NVIDIA RTX A6000, Re-ranking Protocol=Two-Stage VPR2026.04 | 90.1 | 95.7 | 96.9 | — | — | — | — | — | — | — | — | |
| FoLRes=322, Two-stage=true2026.04 | 90.1 | 95.7 | 96.9 | — | — | — | — | — | — | — | — | |
| CricaVPRFeature Dimension=40962024.02 | 90 | 95.4 | 96.4 | — | — | — | — | — | — | — | — | |
| CricaVPRDimension=40962024.05 | 90 | — | — | — | — | — | — | — | — | — | — | |
| CricaVPRtwo-stage=false2024.05 | 90 | 95.4 | 96.4 | — | — | — | — | — | — | — | — | |
| CricaVPRVenue & Year=CVPR’242026.01 | 90 | 95.4 | 96.4 | — | — | — | — | — | — | — | — | |
| CricaVPRDim=4096, Two-stage processing=false2025.09 | 90 | 95.4 | 96.4 | — | — | — | — | — | — | — | — | |
| CricaVPRAttack=None2026.06 | 90 | 95.4 | — | — | — | — | — | — | — | — | — | |
| R2FormerStage=Two-stage2024.02 | 89.7 | 95 | 96.2 | — | — | — | — | — | — | — | — | |
| R2FormerDesc. size (Global)=256, Desc. size (Local)=500 x 131, Memory (GB)=4.7, Latency (ms) Retrieval=8.88, Latency (ms) Reranking=202.372023.11 | 89.7 | 95 | 96.2 | — | — | — | — | — | — | — | — | |
| R2FormerExtrac. (ms)=31, Rerank. (ms)=4002024.05 | 89.7 | 95 | — | — | — | — | — | — | — | — | — | |
| R2Formertwo-stage=true2024.05 | 89.7 | 95 | 96.2 | — | — | — | — | — | — | — | — | |
| R2FormerEnc. Size=ViT-S2024.10 | 89.7 | 95 | 96.2 | — | — | — | — | — | — | — | — | |
| R2FormerDim=N/A, Two-stage processing=true2025.09 | 89.7 | 95 | 96.2 | — | — | — | — | — | — | — | — | |
| EigenPlacesBackbone=ResNet-50, Desc. Dim.=5122023.08 | 89.5 | — | — | — | — | — | — | — | — | — | — | |
| PaVPR baselineEnc. Size=ViT-B2024.10 | 89.5 | 94.5 | 95.8 | — | — | — | — | — | — | — | — | |
| E-LoFTRMatching Time (s)=3.939, Inference Resolution=322 × 322, Device=NVIDIA RTX A6000, Re-ranking Protocol=RANSAC inliers2026.04 | 89.5 | 94.4 | 96.2 | — | — | — | — | — | — | — | — | |
| EigenPlacesDesc. size=2048, Latency (ms)=2.652023.11 | 89.3 | 93.7 | — | — | — | — | — | — | — | — | — | |
| DINOv2 SALADDesc. size=512 + 32, Latency (ms)=2.332023.11 | 89.3 | 94.9 | — | — | — | — | — | — | — | — | — | |
| EigenPlacesDesc. size=2048, Latency (ms)=2.652025.12 | 89.3 | 93.7 | — | — | — | — | — | — | — | — | — |