Image Retrieval on Oxford Revisited (Hard)
58.9mAPCiDeR-FT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CiDeR-FTTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=ArcFace, Fine-tuned=true, Dimension=20482024.04 | 58.9 | — | |
| HesAffNet-HardNet++-HQE+SVDetector=HesAffNet (Hessian + AffNet), Descriptor=HardNet++, Query Expansion=HQE, Spatial Verification=true2017.11 | 53.3 | 72.6 | |
| ResNet50-DELF–HQE+SVBackbone=ResNet-50, Local Feature=DELF, Query Expansion=HQE, Spatial Verification=true2017.11 | 50.3 | 67.2 | |
| HesAff-RootSIFT–HQE+SVDetector=Hessian-Affine, Descriptor=RootSIFT, Query Expansion=HQE, Spatial Verification=true2017.11 | 49.7 | 69.6 | |
| R-ASMK*+SPTrain Set=NC-clean, Backbone=ResNet-50, Pooling=Softmax, Local, Fine-tuned=true2024.04 | 48.5 | — | |
| HesAffNet-HardNet++-HQEDetector=HesAffNet (Hessian + AffNet), Descriptor=HardNet++, Query Expansion=HQE2017.11 | 47.5 | 66.3 | |
| R-ASMK*Train Set=NC-clean, Backbone=ResNet-50, Pooling=Softmax, Local, Fine-tuned=true2024.04 | 45.6 | — | |
| ResNet101-R-MAC+DFSBackbone=ResNet-101, Pooling=R-MAC, Re-ranking=DFS (Global Diffusion)2017.11 | 44.7 | 60.5 | |
| DELF-ASMK*+SPTrain Set=SfM-120k, Backbone=ResNet-50, Pooling=Softmax, Fine-tuned=false2024.04 | 43.1 | — | |
| CiDeRTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=ArcFace, Fine-tuned=false, Dimension=20482024.04 | 42.4 | — | |
| HesAff-RootSIFT-HQEDetector=Hessian-Affine, Descriptor=RootSIFT, Query Expansion=HQE2017.11 | 41.3 | 60 | |
| ResNet101-GeM+αQEBackbone=ResNet-101, Pooling=GeM, Query Expansion=αQE2017.11 | 40.7 | 54.9 | |
| AGEMTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=Siamese, Fine-tuned=true, Dimension=20482024.04 | 40.7 | — | |
| ResNet101-GeM+DFSBackbone=ResNet-101, Pooling=GeM, Re-ranking=DFS (Global Diffusion)2017.11 | 40.5 | 54.4 | |
| GLAMTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=ArcFace, Fine-tuned=true, Dimension=5122024.04 | 39.5 | — | |
| Radenovic et al.Train Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=Siamese, Fine-tuned=true, Dimension=20482024.04 | 38.5 | — | |
| HesAff-rSIFT-ASMK*+SPTrain Set=SfM-120k, Backbone=ResNet-50, Fine-tuned=false2024.04 | 36.7 | — | |
| SOLARTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM, Loss=Triplet, SOS, Fine-tuned=true, Dimension=20482024.04 | 27.1 | — | |
| DIRTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=RMAC, Loss=Triplet, Fine-tuned=true, Dimension=20482024.04 | 25.5 | — | |
| DINOArch.=ViT-S/16, Pretrain Dataset=GLDv22021.04 | 24.3 | — | |
| RN101+R-MACPretrain=Supervised, Pretrain Dataset=ImNet2021.04 | 18.5 | — | |
| DOLGTrain Set=SfM-120k, Backbone=ResNet-101, Pooling=GeM,GAP, Loss=ArcFace, Fine-tuned=true, Dimension=5122024.04 | 18.1 | — | |
| MSNBackbone=ViT-Small/162024.05 | 15.1 | — | |
| iBOT-vMFBackbone=ViT-Base/162024.05 | 13.8 | — | |
| DINOArch.=ViT-S/16, Pretrain Dataset=ImNet2021.04 | 13.7 | — | |
| iBOTBackbone=ViT-Base/162024.05 | 13.4 | — | |
| DINOBackbone=ViT-Small/162024.05 | 13 | — | |
| DINO-vMFBackbone=ViT-Small/162024.05 | 12.6 | — | |
| DINO-vMFBackbone=ViT-Base/162024.05 | 12.5 | — | |
| MSNBackbone=ViT-Base/162024.05 | 11.8 | — | |
| DINOArch.=ResNet-50, Pretrain Dataset=ImNet2021.04 | 11.1 | — | |
| DINOBackbone=ViT-Base/162024.05 | 10.8 | — | |
| ViT-S/16Pretrain=Supervised, Pretrain Dataset=ImNet2021.04 | 8.9 | — |