Image Retrieval on CUB-200 2011 (test)
87.2Recall@1LaFG
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LaFGArch=ViT2025.12 | 87.2 | 92.4 | 95.2 | 97 | — | — | — | — | — | — | — | — | |
| VPTSP-GArch=ViT2025.12 | 86.6 | 91.7 | 94.8 | — | — | — | — | — | — | — | — | — | |
| DDMLArch=ViT2025.12 | 86 | 91.7 | 95.2 | 96.8 | — | — | — | — | — | — | — | — | |
| SEEArch=ViT2025.12 | 85.8 | 91.4 | 94.6 | — | — | — | — | — | — | — | — | — | |
| HIERArch=ViT2025.12 | 85.7 | 91.3 | 94.4 | — | — | — | — | — | — | — | — | — | |
| HypViTArch=ViT2025.12 | 85.6 | 91.4 | 94.8 | 96.7 | — | — | — | — | — | — | — | — | |
| DPHMArch=ViT2025.12 | 85.5 | 91.3 | 94.6 | 96.6 | — | — | — | — | — | — | — | — | |
| DVAArch=ViT2025.12 | 84.9 | 90.6 | 94.5 | 96.7 | — | — | — | — | — | — | — | — | |
| DFML-PAArch=ViT2025.12 | 79.1 | 86.8 | — | — | — | — | — | — | — | — | — | — | |
| ROADMAPdim=384, Category=DeiT, Backbone=DeiT2021.10 | 77.4 | 85.5 | 91.4 | 95 | — | — | — | — | — | — | — | — | |
| LM-MetricArch=ViT2025.12 | 77.1 | — | — | — | — | — | — | — | — | — | — | — | |
| DIMLArch=ViT2025.12 | 76.7 | — | — | — | — | — | — | — | — | — | — | — | |
| IRTRdim=384, Category=DeiT, Backbone=DeiT2021.10 | 76.6 | 85 | 91.1 | 94.3 | — | — | — | — | — | — | — | — | |
| CCL*Backbone=ResNet-50, Embedding Size=512, margin (m)=nonzero, Input Size=Larger2023.08 | 73.5 | 81.9 | 87.8 | — | — | — | — | — | — | — | — | — | |
| ProxyNCA++Arch=R50, Emb=20482020.04 | 72.2 | 82 | 89.2 | 93.5 | — | — | — | 75.8 | — | — | — | — | |
| ProxyNCA++dim=2048, Category=Classification, Backbone=ResNet-502021.10 | 72.2 | 82 | 89.2 | 93.5 | — | — | — | — | — | — | — | — | |
| PNCA++Arch=R502025.12 | 72.2 | 82 | 89.2 | 93.5 | — | — | — | — | — | — | — | — | |
| ProxyAnchor-AVSLSetting=512R2022.03 | 71.9 | 81.7 | 88.1 | 93.2 | — | — | — | — | — | — | — | — | |
| CCLBackbone=ResNet-50, Embedding Size=512, margin (m)=0, Input Size=Larger2023.08 | 71.8 | 80.8 | 87.8 | — | — | — | — | — | — | — | — | — | |
| HISTArch=R502025.12 | 71.4 | 81.1 | 88.1 | — | — | — | — | — | — | — | — | — | |
| Proxy-AnchorEmbedding size=512, Backbone=BN-Inception, Image Size=256x2562020.03 | 71.1 | 80.4 | 87.4 | 92.5 | — | — | — | — | — | — | — | — | |
| ProxyAnchor+Backbone=Inception, Embedding Size=5122023.08 | 71.1 | 80.4 | 87.4 | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet-50, Embedding Size=512, margin (m)=0, lambda=0, Input Size=Larger2023.08 | 71.1 | 80.8 | 88.2 | — | — | — | — | — | — | — | — | — | |
| IDMLArch=R502025.12 | 70.7 | 80.2 | — | — | — | — | — | — | — | — | — | — | |
| HSEArch=R502025.12 | 70.6 | 80.1 | 87.1 | — | — | — | — | — | — | — | — | — | |
| NIRArch=R502025.12 | 70.5 | 80.6 | — | — | — | — | — | — | — | — | — | — | |
| MetrixBackbone=ResNet-50, Embedding Size=5122023.08 | 70.4 | 80.6 | 88.7 | — | — | — | — | — | — | — | — | — | |
| ProxyNCA++Arch=R50, Emb=10242020.04 | 70.2 | 80.7 | 88 | 93 | — | — | — | 74.2 | — | — | — | — | |
| CBMLArch=R502025.12 | 69.9 | 80.4 | 87.2 | 92.5 | — | — | — | — | — | — | — | — | |
| ProxyAnchorSetting=512R2022.03 | 69.7 | 80 | 87 | 92.4 | — | — | — | — | — | — | — | — | |
| Cross-Entropydim=2048, Category=Classification, Backbone=ResNet-502021.10 | 69.2 | 79.2 | 86.9 | 91.6 | — | — | — | — | — | — | — | — | |
| CEBackbone=ResNet-50, Embedding Size=20482023.08 | 69.2 | 79.2 | 86.9 | — | — | — | — | — | — | — | — | — | |
| ProxyNCA++ (-max, -fast)Arch=R50, Emb=20482020.04 | 69.1 | 79.6 | 87.3 | 92.7 | — | — | — | 73.3 | — | — | — | — | |
| ProxyNCA++Arch=R50, Emb=5122020.04 | 69 | 79.8 | 87.3 | 92.7 | — | — | — | 73.9 | — | — | — | — | |
| ProxyNCA++dim=512, Category=Classification, Backbone=ResNet-502021.10 | 69 | 79.8 | 87.3 | 92.7 | — | — | — | — | — | — | — | — | |
| ProxyNCA++Backbone=ResNet-50, Embedding Size=5122023.08 | 69 | 79.8 | 87.3 | — | — | — | — | — | — | — | — | — | |
| SECdim=512, Category=Metric learning, Backbone=ResNet-502021.10 | 68.8 | 79.4 | 87.2 | 92.5 | — | — | — | — | — | — | — | — | |
| Margin-AVSLSetting=512R2022.03 | 68.8 | 79.2 | 87.3 | 92.7 | — | — | — | — | — | — | — | — | |
| DRMLSetting=512R2022.03 | 68.7 | 78.6 | 86.3 | 91.6 | — | — | — | — | — | — | — | — | |
| ROADMAPdim=512, Category=AP loss, Backbone=ResNet-502021.10 | 68.5 | 78.7 | 86.6 | 91.9 | — | — | — | — | — | — | — | — | |
| ROADMAPBackbone=ResNet-50, Embedding Size=512, Input Size=Larger2023.08 | 68.5 | 78.7 | 86.6 | — | — | — | — | — | — | — | — | — | |
| Proxy-AnchorEmbedding size=512, Backbone=BN-Inception, Image Size=224x2242020.03 | 68.4 | 79.2 | 86.8 | 91.6 | — | — | — | — | — | — | — | — | |
| DCMLSetting=512R2022.03 | 68.4 | 77.9 | 86.1 | 91.7 | — | — | — | — | — | — | — | — | |
| DIMLSetting=512R2022.03 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | |
| ArcFaceBackbone=ResNet-50, Embedding Size=5122023.08 | 67.5 | — | — | — | — | — | — | — | — | — | — | — | |
| CosFaceBackbone=ResNet-50, Embedding Size=5122023.08 | 67.3 | — | — | — | — | — | — | — | — | — | — | — | |
| contrastive loss + HORDEBackbone=BN-Inception, Evaluation Protocol=high-order moments concatenation + PCA2019.08 | 66.8 | 77.4 | 85.1 | 91 | 94.8 | 97.3 | — | — | — | — | — | — | |
| HORDEArch=I3, Emb=5122020.04 | 66.8 | 77.4 | 85.1 | 91 | — | — | — | — | — | — | — | — | |
| HORDEdim=512, Category=Metric learning, Backbone=ResNet-502021.10 | 66.8 | 77.4 | 85.1 | 91 | — | — | — | — | — | — | — | — | |
| HORDEBackbone=Inception, Embedding Size=512, Input Size=Larger2023.08 | 66.8 | 77.4 | 85.1 | — | — | — | — | — | — | — | — | — | |
| CircleLossEmbedding size=5122020.02 | 66.7 | 77.4 | 86.2 | 91.2 | — | — | — | — | — | — | — | — | |
| CircleSetting=512BN2022.03 | 66.7 | 77.4 | 86.2 | 91.2 | — | — | — | — | — | — | — | — | |
| CircleLossBackbone=ResNet-50, Embedding Size=5122023.08 | 66.7 | 77.4 | 86.2 | — | — | — | — | — | — | — | — | — | |
| ProxyGMLdim=512, Category=Classification, Backbone=ResNet-502021.10 | 66.6 | 77.6 | 86.4 | — | — | — | — | — | — | — | — | — | |
| ProxyGMLBackbone=Inception, Embedding Size=5122023.08 | 66.6 | 77.6 | 86.4 | — | — | — | — | — | — | — | — | — | |
| contrastive loss + HORDEBackbone=BN-Inception, Evaluation Protocol=standard2019.08 | 66.3 | 76.7 | 84.7 | 90.6 | 94.5 | 96.7 | — | — | — | — | — | — | |
| Contra+HORDEEmbedding size=512, Backbone=BN-Inception, Image Size=224x2242020.03 | 66.3 | 76.7 | 84.7 | 90.6 | — | — | — | — | — | — | — | — | |
| MIC+marginDim=128, Backbone=ResNet-502019.09 | 66.1 | 76.8 | 85.6 | — | — | — | — | 69.7 | — | — | — | — | |
| MICdim=512, Category=Metric learning, Backbone=ResNet-502021.10 | 66.1 | 76.8 | 85.6 | — | — | — | — | — | — | — | — | — | |
| XBMdim=128, Category=Metric learning, Backbone=ResNet-502021.10 | 65.8 | 75.9 | 84 | 89.9 | — | — | — | — | — | — | — | — | |
| XBMBackbone=Inception, Embedding Size=5122023.08 | 65.8 | 75.9 | 84 | — | — | — | — | — | — | — | — | — | |
| Multi-similarity lossBackbone=BN-Inception, Evaluation Protocol=standard2019.08 | 65.7 | 77 | 86.3 | 91.2 | 95 | 97.3 | — | — | — | — | — | — | |
| MSEmbedding size=512, Backbone=BN-Inception, Image Size=224x2242020.03 | 65.7 | 77 | 86.3 | 91.2 | — | — | — | — | — | — | — | — | |
| Multi-SimiEmbedding size=5122020.02 | 65.7 | 77 | 86.3 | 91.2 | — | — | — | — | — | — | — | — | |
| MSArch=I3, Emb=5122020.04 | 65.7 | 77 | 86.3 | 91.2 | — | — | — | — | — | — | — | — | |
| MSdim=512, Category=Metric learning, Backbone=ResNet-502021.10 | 65.7 | 77 | 86.3 | 91.2 | — | — | — | — | — | — | — | — | |
| MSSetting=512BN2022.03 | 65.7 | 77 | 86.3 | 91.2 | — | — | — | — | — | — | — | — | |
| MSBackbone=Inception, Embedding Size=5122023.08 | 65.7 | 77 | 86.3 | — | — | — | — | — | — | — | — | — | |
| MarginSetting=512R2022.03 | 65.6 | 75.9 | 84.3 | 90.8 | — | — | — | — | — | — | — | — | |
| A-BIER2018.01 | 65.5 | 75.8 | 83.9 | 90.2 | 94.2 | 97.1 | — | — | — | — | — | — | |
| SoftTripleEmbedding size=512, Backbone=BN-Inception, Image Size=224x2242020.03 | 65.4 | 76.4 | 84.5 | 90.4 | — | — | — | — | — | — | — | — | |
| SoftTripleSetting=512BN2022.03 | 65.4 | 76.4 | 84.5 | 91.6 | — | — | — | — | — | — | — | — | |
| SoftTripleBackbone=ResNet-50, Embedding Size=5122023.08 | 65.4 | 76.4 | 84.5 | — | — | — | — | — | — | — | — | — | |
| NormSoftmaxEmbedding dimension=2048, Net=ResNet-502018.11 | 65.3 | 76.7 | 85.4 | 91.8 | — | — | — | — | — | — | — | — | |
| NormSoftMaxArch=R50, Emb=20482020.04 | 65.3 | 76.7 | 85.4 | 91.8 | — | — | — | — | — | — | — | — | |
| NSoftmaxdim=2048, Category=Classification, Backbone=ResNet-502021.10 | 65.3 | 76.7 | 85.4 | 91.8 | — | — | — | — | — | — | — | — | |
| NSoftmaxBackbone=ResNet-50, Embedding Size=20482023.08 | 65.3 | 76.7 | 85.4 | — | — | — | — | — | — | — | — | — | |
| A-BIERensemble_size=Learner-32018.01 | 64 | 74.3 | 83.1 | 89.2 | 94.1 | 96.9 | — | — | — | — | — | — | |
| BlackBoxdim=512, Category=AP loss, Backbone=ResNet-502021.10 | 64 | 75.3 | 84.1 | 90.6 | — | — | — | — | — | — | — | — | |
| BlackBoxBackbone=ResNet-50, Embedding Size=5122023.08 | 64 | 75.3 | 84.1 | — | — | — | — | — | — | — | — | — | |
| Margin2018.01 | 63.9 | 75.3 | 84.4 | 90.6 | 94.8 | — | — | — | — | — | — | — | |
| MarginInput Image Transformation=Cropped Images2017.06 | 63.9 | 75.3 | 84.4 | 90.6 | 94.8 | — | 69.8 | — | — | — | — | — | |
| ABE + HORDEBackbone=GoogleNet, Evaluation Protocol=high-order moments concatenation + PCA2019.08 | 63.9 | 75.7 | 84.4 | 91.2 | 95.3 | 97.6 | — | — | — | — | — | — | |
| DREMLEmbedding dimension=576, Net=ResNet-182018.11 | 63.9 | 75 | 83.1 | 89.7 | — | — | — | — | — | — | — | — | |
| DREMLDim=92162019.09 | 63.9 | 75 | 83.1 | — | — | — | — | 67.8 | — | — | — | — | |
| BIER2018.01 | 63.7 | 74 | 82.5 | 89.3 | 93.8 | 96.8 | — | — | — | — | — | — | |
| MarginInput Image Transformation=Original Images2017.06 | 63.6 | 74.4 | 83.1 | 90 | 94.2 | — | 69 | — | — | — | — | — | |
| MarginEmbedding Size=128, Backbone=ResNet-502018.04 | 63.6 | 74.4 | 83.1 | 90 | — | — | — | — | — | — | — | — | |
| MarginEmbedding dimension=128, Net=ResNet-502018.11 | 63.6 | 74.4 | 83.1 | 90 | — | — | — | — | — | — | — | — | |
| MarginDim=1282019.09 | 63.6 | 74.4 | 83.1 | — | — | — | — | 69 | — | — | — | — | |
| MarginEmbedding size=128, Backbone=ResNet50, Image Size=224x2242020.03 | 63.6 | 74.4 | 83.1 | 90 | — | — | — | — | — | — | — | — | |
| MarginArch=R50, Emb=1282020.04 | 63.6 | 74.4 | 83.1 | 90 | — | — | — | 69 | — | — | — | — | |
| Triplet SHdim=512, Category=Metric learning, Backbone=ResNet-502021.10 | 63.6 | 74.4 | 83.1 | 90 | — | — | — | — | — | — | — | — | |
| Triplet-SHBackbone=ResNet-50, Embedding Size=5122023.08 | 63.6 | 74.4 | 83.1 | — | — | — | — | — | — | — | — | — | |
| NormSoftmaxEmbedding dimension=2048, Net=ResNet-50, Output format=bits2018.11 | 63.3 | 75.2 | 84.3 | 91 | — | — | — | — | — | — | — | — | |
| Margin*Dim=128, Backbone=ResNet-502019.09 | 62.9 | 74.1 | 82.9 | — | — | — | — | 66.3 | — | — | — | — | |
| BIERensemble_size=Learner-32018.01 | 62.8 | 73.5 | 81.9 | 89 | 93.7 | 96.7 | — | — | — | — | — | — | |
| ABE + HORDEBackbone=GoogleNet, Evaluation Protocol=standard2019.08 | 62.7 | 74.3 | 83.4 | 90.2 | 94.6 | 96.9 | — | — | — | — | — | — | |
| NormSoftmaxEmbedding dimension=1024, Net=BNInception2018.11 | 62.2 | 73.9 | 82.7 | 89.4 | — | — | — | — | — | — | — | — | |
| Proxy-AnchorEmbedding size=64, Backbone=BN-Inception, Image Size=224x2242020.03 | 61.7 | 73 | 81.8 | 88.8 | — | — | — | — | — | — | — | — |