Image Retrieval on Oxford 5k
94.7mAPDeep Image Retrieval (with QE and DBA)
Evaluation results
| Method | Links | |
|---|---|---|
| Deep Image Retrieval (with QE and DBA)Dim.=2048, Category=Matching / Spatial verif. / QE, Query Expansion (QE)=true, Database Augmentation (DBA)=true2016.10 | 94.7 | |
| CiDeR-FTTraining Set=SfM-120k, Fine-tuned=true2024.04 | 92.6 | |
| IME layerDimensions=2048, Method category=Dimension reduction and manifold learning2017.07 | 92 | |
| IME layerDimensions=2048, Method category=Search reranking2017.07 | 92 | |
| IME layerDimensions=1024, Method category=Dimension reduction and manifold learning2017.07 | 91.8 | |
| IME layerDimensions=512, Method category=Dimension reduction and manifold learning2017.07 | 91.2 | |
| GLAMTraining Set=RGLDv2-clean, Fine-tuned=false2024.04 | 90.9 | |
| CiDeR-FTTraining Set=RGLDv2-clean, Fine-tuned=true2024.04 | 90.9 | |
| Deep Image Retrieval (with QE)Dim.=2048, Category=Matching / Spatial verif. / QE, Query Expansion (QE)=true2016.10 | 90.6 | |
| SOLARTraining Set=RGLDv2-clean, Fine-tuned=true2024.04 | 90.6 | |
| CiDeRTraining Set=SfM-120k, Fine-tuned=false2024.04 | 89.9 | |
| CiDeRTraining Set=RGLDv2-clean, Fine-tuned=false2024.04 | 89.8 | |
| GLAMTraining Set=SfM-120k, Fine-tuned=false2024.04 | 89.7 | |
| QEDimensions=2048, Method category=Search reranking2017.07 | 89.6 | |
| HesAffNet-HardNet++-HQEVocabulary Size=65k, Assignment=Multiple Assignments (MA), Spatial Verification=true, Query Expansion=true2017.11 | 89.5 | |
| Tolias & JégouRetrieval Protocol=Matching / Spatial verif. / QE2016.04 | 89.4 | |
| Tolias and Jégou (2015)Category=Matching / Spatial verif. / QE2016.10 | 89.4 | |
| Deep Image Retrieval + QE (VGG16)Dim.=512, Retrieval Protocol=Matching / Spatial verif. / QE, Backbone=VGG162016.04 | 89.1 | |
| SCSMDimensions=2048, Method category=Search reranking2017.07 | 89.1 | |
| Deep Image Retrieval + QE (ResNet50)Dim.=2048, Retrieval Protocol=Matching / Spatial verif. / QE, Backbone=ResNet502016.04 | 89 | |
| HesAff-HardNet++-HQEVocabulary Size=65k, Assignment=Multiple Assignments (MA), Spatial Verification=true, Query Expansion=true2017.11 | 88.3 | |
| DOLGTraining Set=RGLDv2-clean, Fine-tuned=false2024.04 | 88.3 | |
| HesAff-RootSIFT-HQEVocabulary Size=65k, Assignment=Multiple Assignments (MA), Spatial Verification=true, Query Expansion=true2017.11 | 88 | |
| HesAffNet-HardNet++-HQEVocabulary Size=65k, Assignment=Single Assignment (SA), Spatial Verification=true, Query Expansion=true2017.11 | 87.9 | |
| Filip et al.Training Set=SfM-120k, Fine-tuned=true2024.04 | 87.8 | |
| DiffusionDimensions=2048, Method category=Search reranking2017.07 | 87.1 | |
| Tolias et al. (QE)Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 86.9 | |
| Tolias et al (2015)Category=Matching / Spatial verif. / QE2016.10 | 86.9 | |
| HesAff-HardNet++-HQEVocabulary Size=65k, Assignment=Single Assignment (SA), Spatial Verification=true, Query Expansion=true2017.11 | 86.8 | |
| Deep Image Retrieval (Ours)Dim.=2048, Category=Global descriptors2016.10 | 86.1 | |
| Yokoo et al.Training Set=RGLDv2-clean, Fine-tuned=false2024.04 | 86.1 | |
| HesAff-RootSIFT-HQEVocabulary Size=65k, Assignment=Single Assignment (SA), Spatial Verification=true, Query Expansion=true2017.11 | 85.3 | |
| Radenovic et al.Dim.=512, Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 85 | |
| Radenovic et al (2016)Dim.=512, Category=Matching / Spatial verif. / QE2016.10 | 85 | |
| Mikulik et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 84.9 | |
| Mikulik et al (2013)Category=Matching / Spatial verif. / QE2016.10 | 84.9 | |
| HesAff-SIFT-BoW-fVocabVocabulary Size=16M, Assignment=Multiple Assignments (MA), Spatial Verification=true, Query Expansion=true2017.11 | 84.9 | |
| Deep Image Retrieval (ResNet50)Dim.=2048, Retrieval Protocol=Global descriptors, Backbone=ResNet502016.04 | 84.5 | |
| Deng et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 84.3 | |
| Deng et al (2013)Category=Matching / Spatial verif. / QE2016.10 | 84.3 | |
| R-MACDimensions=2048, Method category=Original representations2017.07 | 83.9 | |
| Deep Image Retrieval (VGG16)Dim.=512, Retrieval Protocol=Global descriptors, Backbone=VGG162016.04 | 83.1 | |
| Chum et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 82.7 | |
| Chum et al (2011)Category=Matching / Spatial verif. / QE2016.10 | 82.7 | |
| ICADimensions=2048, Method category=Dimension reduction and manifold learning2017.07 | 82.7 | |
| PCADimensions=2048, Method category=Dimension reduction and manifold learning2017.07 | 82.6 | |
| ICADimensions=1024, Method category=Dimension reduction and manifold learning2017.07 | 81.6 | |
| Danfeng et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 81.4 | |
| Danfeng et al (2011)Category=Matching / Spatial verif. / QE2016.10 | 81.4 | |
| TokenTraining Set=RGLDv2-clean, Fine-tuned=false2024.04 | 81.2 | |
| PCADimensions=1024, Method category=Dimension reduction and manifold learning2017.07 | 80.8 | |
| Radenovic et al.Dim.=512, Retrieval Protocol=Global descriptors2016.04 | 79.7 | |
| Radenovic et al (2016)Dim.=512, Category=Global descriptors2016.10 | 79.7 | |
| Azizpour et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 79 | |
| Azizpour et al (2015)Category=Matching / Spatial verif. / QE2016.10 | 79 | |
| DIRTraining Set=SfM-120k, Fine-tuned=false2024.04 | 79 | |
| GeMTraining Set=SfM-120k, Fine-tuned=false2024.04 | 79 | |
| Tolias et al (2016) (ResNet101)Dim.=2048, Category=Matching / Spatial verif. / QE, reimplemented=true2016.10 | 78.9 | |
| SOLARTraining Set=SfM-120k, Fine-tuned=true2024.04 | 78.5 | |
| PCADimensions=512, Method category=Dimension reduction and manifold learning2017.07 | 78.2 | |
| IsoMapDimensions=1024, Method category=Dimension reduction and manifold learning2017.07 | 78.1 | |
| IsoMapDimensions=2048, Method category=Dimension reduction and manifold learning2017.07 | 77.9 | |
| Tao et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 77.8 | |
| Tao et al (2014)Category=Matching / Spatial verif. / QE2016.10 | 77.8 | |
| IsoMapDimensions=512, Method category=Dimension reduction and manifold learning2017.07 | 77.8 | |
| R-MACDimensions=512, Method category=Original representations2017.07 | 77.7 | |
| ICADimensions=512, Method category=Dimension reduction and manifold learning2017.07 | 77.5 | |
| Tolias et al. (R-MAC)Dim.=512, Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 77.3 | |
| Tolias et al (2016)Dim.=512, Category=Matching / Spatial verif. / QE2016.10 | 77.3 | |
| Shen et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 75.2 | |
| Shen et al (2014)Category=Matching / Spatial verif. / QE2016.10 | 75.2 | |
| HesAff-SIFT-BoW-fVocabVocabulary Size=16M, Assignment=Single Assignment (SA), Spatial Verification=true, Query Expansion=true2017.11 | 74 | |
| Xinchao et al.Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 73.7 | |
| Li et al (2015)Category=Matching / Spatial verif. / QE2016.10 | 73.7 | |
| DOLGTraining Set=SfM-120k, Fine-tuned=false2024.04 | 72.8 | |
| Kalantidis et al.Dim.=512, Retrieval Protocol=Matching / Spatial verif. / QE2016.04 | 72.2 | |
| Kalantidis et al (2016)Dim.=512, Category=Matching / Spatial verif. / QE2016.10 | 72.2 | |
| Arandjelovic et al.Dim.=4096, Retrieval Protocol=Global descriptors2016.04 | 71.6 | |
| Arandjelovic et al (2016)Dim.=4096, Category=Global descriptors2016.10 | 71.6 | |
| Tolias et al (2016) (ResNet101)Dim.=2048, Category=Global descriptors, reimplemented=true2016.10 | 69.4 | |
| Kalantidis et al.Dim.=512, Retrieval Protocol=Global descriptors2016.04 | 68.2 | |
| Kalantidis et al (2016)Dim.=512, Category=Global descriptors2016.10 | 68.2 | |
| CroWDimensions=512, Method category=Original representations2017.07 | 68.2 | |
| Tolias et al.Dim.=512, Retrieval Protocol=Global descriptors2016.04 | 66.9 | |
| Tolias et al (2016)Dim.=512, Category=Global descriptors2016.10 | 66.9 | |
| LLEDimensions=512, Method category=Dimension reduction and manifold learning2017.07 | 64 | |
| Ng et al.Dim.=128, Retrieval Protocol=Global descriptors, Query Region=Full image2016.04 | 59.3 | |
| Ng et al (2015)Dim.=128, Category=Global descriptors, full image as query=true2016.10 | 59.3 | |
| LLEDimensions=1024, Method category=Dimension reduction and manifold learning2017.07 | 58.8 | |
| Paulin et al.Dim.=256K, Retrieval Protocol=Global descriptors2016.04 | 56.5 | |
| Paulin et al (2015)Dim.=256K, Category=Global descriptors2016.10 | 56.5 | |
| Jégou & ZissermanDim.=1024, Retrieval Protocol=Global descriptors2016.04 | 56 | |
| Jégou and Zisserman (2014)Dim.=1024, Category=Global descriptors2016.10 | 56 | |
| Babenko et al.Dim.=128, Retrieval Protocol=Global descriptors, Query Region=Full image2016.04 | 55.7 | |
| Babenko et al (2014)Dim.=128, Category=Global descriptors, full image as query=true2016.10 | 55.7 | |
| Babenko & LempitskyDim.=256, Retrieval Protocol=Global descriptors2016.04 | 53.1 | |
| Babenko and Lempitsky (2015)Dim.=256, Category=Global descriptors2016.10 | 53.1 | |
| LLEDimensions=2048, Method category=Dimension reduction and manifold learning2017.07 | 51.7 | |
| Jégou & ZissermanDim.=128, Retrieval Protocol=Global descriptors2016.04 | 43.3 | |
| Jégou and Zisserman (2014)Dim.=128, Category=Global descriptors2016.10 | 43.3 |