Image Retrieval on CUB 200 2011
89.1Recall@1GAPan
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GAPanArch=ViT2026.05 | 89.1 | 93.5 | 95.3 | 97.1 | — | — | |
| Potential FieldBackbone=ViT, Embedding Dimension=3842024.05 | 87.8 | 92.6 | 95.7 | — | — | — | |
| Potential FieldBackbone=ViT, Embedding Dimension=384, Variant=PF, Training Setting=Original Paper Reported (†)2026.01 | 87.8 | 92.6 | 95.7 | — | — | — | |
| LaFGArch=ViT2026.05 | 87.2 | 92.4 | 95.2 | 97 | — | — | |
| VPTSP-GICLRArch=ViT2026.05 | 86.6 | 91.7 | 94.8 | — | — | — | |
| DDMLArch=ViT2026.05 | 86 | 91.7 | 95.2 | 96.8 | — | — | |
| SEEIArch=ViT2026.05 | 85.8 | 91.4 | 94.6 | — | — | — | |
| HIERBackbone=ViT, Embedding Dimension=3842024.05 | 85.7 | 91.3 | 94.4 | — | — | — | |
| HIERBackbone=ViT, Embedding Dimension=3842026.01 | 85.7 | 91.3 | 94.4 | — | — | — | |
| HIERArch=ViT2026.05 | 85.7 | 91.3 | 94.4 | — | — | — | |
| HypBackbone=ViT, Embedding Dimension=3842024.05 | 85.6 | 91.4 | 94.8 | — | — | — | |
| HypBackbone=ViT, Embedding Dimension=3842026.01 | 85.6 | 91.4 | 94.8 | — | — | — | |
| HypViTArch=ViT2026.05 | 85.6 | 91.4 | 94.8 | 96.7 | — | — | |
| DPHMArch=ViT2026.05 | 85.5 | 91.3 | 94.6 | 96.6 | — | — | |
| DVAArch=ViT2026.05 | 84.9 | 90.6 | 94.5 | 96.7 | — | — | |
| BLenDeR - PABackbone=ViT, Embedding Dimension=384, Base Method=Proxy Anchor2026.01 | 84.4 | 90.7 | 94.2 | — | — | — | |
| Proxy AnchorBackbone=ViT, Embedding Dimension=384, Variant=PA, Training Setting=Replicated (*)2026.01 | 84.1 | 90.1 | 94.3 | — | — | — | |
| BLenDeR - PFBackbone=ViT, Embedding Dimension=384, Base Method=Potential Field2026.01 | 84 | 90.5 | 94.1 | — | — | — | |
| Potential FieldBackbone=DINO, Embedding Dimension=3842024.05 | 83.1 | 89.3 | 94.2 | — | — | — | |
| ViT-SBackbone=ViT, Embedding Dimension=3842024.05 | 83.1 | 90.4 | 94.4 | — | — | — | |
| Potential FieldBackbone=DINO, Embedding Dimension=384, Variant=PF, Training Setting=Original Paper Reported (†)2026.01 | 83.1 | 89.3 | 94.2 | — | — | — | |
| ViT-SBackbone=ViT, Embedding Dimension=3842026.01 | 83.1 | 90.4 | 94.4 | — | — | — | |
| Potential FieldBackbone=ViT, Embedding Dimension=384, Variant=PF, Training Setting=Replicated (*)2026.01 | 83.1 | 90 | 94 | — | — | — | |
| BLenDeR - PABackbone=DINO, Embedding Dimension=384, Base Method=Proxy Anchor2026.01 | 81.8 | 88.9 | 93 | — | — | — | |
| BLenDeR - PFBackbone=DINO, Embedding Dimension=384, Base Method=Potential Field2026.01 | 81.4 | 88.1 | 92.8 | — | — | — | |
| Proxy AnchorBackbone=DINO, Embedding Dimension=384, Variant=PA, Training Setting=Replicated (*)2026.01 | 81.2 | 88.8 | 93.1 | — | — | — | |
| HIERBackbone=DINO, Embedding Dimension=3842024.05 | 81.1 | 88.2 | 93.3 | — | — | — | |
| HIERBackbone=DINO, Embedding Dimension=3842026.01 | 81.1 | 88.2 | 93.3 | — | — | — | |
| HypBackbone=DINO, Embedding Dimension=3842024.05 | 80.9 | 87.6 | 92.4 | — | — | — | |
| HypBackbone=DINO, Embedding Dimension=3842026.01 | 80.9 | 87.6 | 92.4 | — | — | — | |
| Potential FieldBackbone=DINO, Embedding Dimension=384, Variant=PF, Training Setting=Replicated (*)2026.01 | 80.9 | 88.2 | 92.7 | — | — | — | |
| DFML-PACArch=ViT2026.05 | 79.1 | 86.8 | — | — | — | — | |
| BLenDeR - PFBackbone=ResNet50, Embedding Dimension=512, Base Method=Potential Field2026.01 | 77 | 85.7 | 91.1 | — | — | — | |
| DIMLArch=ViT2026.05 | 76.7 | — | — | — | — | — | |
| BLenDeR - PABackbone=ResNet50, Embedding Dimension=512, Base Method=Proxy Anchor2026.01 | 74.6 | 83.7 | 89.7 | — | — | — | |
| Potential FieldBackbone=ResNet50, Embedding Dimension=5122024.05 | 73.4 | 82.4 | 88.8 | — | — | — | |
| Potential FieldBackbone=ResNet50, Embedding Dimension=512, Variant=PF, Training Setting=Original Paper Reported (†)2026.01 | 73.4 | 82.4 | 88.8 | — | — | — | |
| Potential FieldBackbone=ResNet50, Embedding Dimension=512, Variant=PF, Training Setting=Replicated (*)2026.01 | 73.3 | 83 | 89 | — | — | — | |
| Proxy AnchorBackbone=ResNet50, Embedding Dimension=512, Variant=PA, Training Setting=Replicated (*)2026.01 | 72.7 | 82.1 | 88.7 | — | — | — | |
| PNCA++Arch=R502026.05 | 72.2 | 82 | 89.2 | 93.5 | — | — | |
| Potential FieldBackbone=BN Inception, Embedding Dimension=5122024.05 | 71.5 | 81.2 | 88.3 | — | — | — | |
| HISTBackbone=ResNet50, Embedding Dimension=5122024.05 | 71.4 | 81.1 | 88.1 | — | — | — | |
| HISTBackbone=ResNet50, Embedding Dimension=5122026.01 | 71.4 | 81.1 | 88.1 | — | — | — | |
| HISTArch=R502026.05 | 71.4 | 81.1 | 88.1 | — | — | — | |
| DINOBackbone=DINO, Embedding Dimension=3842024.05 | 70.8 | 81.1 | 88.8 | — | — | — | |
| DINOBackbone=DINO, Embedding Dimension=3842026.01 | 70.8 | 81.1 | 88.8 | — | — | — | |
| IDMLArch=R502026.05 | 70.7 | 80.2 | — | — | — | — | |
| HSE-PABackbone=ResNet50, Embedding Dimension=5122024.05 | 70.6 | 80.1 | 87.1 | — | — | — | |
| HSE-PABackbone=ResNet50, Embedding Dimension=5122026.01 | 70.6 | 80.1 | 87.1 | — | — | — | |
| HSEArch=R502026.05 | 70.6 | 80.1 | 87.1 | — | — | — | |
| NIRArch=R502026.05 | 70.5 | 80.6 | — | — | — | — | |
| HIERBackbone=ResNet50, Embedding Dimension=5122024.05 | 70.1 | 79.4 | 86.9 | — | — | — | |
| HIERBackbone=ResNet50, Embedding Dimension=5122026.01 | 70.1 | 79.4 | 86.9 | — | — | — | |
| CBMLArch=R502026.05 | 69.9 | 80.4 | 87.2 | 92.5 | — | — | |
| Proxy AnchorBackbone=ResNet50, Embedding Dimension=5122024.05 | 69.7 | 80 | 87 | — | — | — | |
| HISTBackbone=BN Inception, Embedding Dimension=5122024.05 | 69.7 | 80 | 87.3 | — | — | — | |
| Proxy AnchorBackbone=ResNet50, Embedding Dimension=512, Variant=PA2026.01 | 69.7 | 80 | 87 | — | — | — | |
| PA+niVBackbone=BN Inception, Embedding Dimension=5122024.05 | 69.5 | 80 | 86.4 | — | — | — | |
| DFML-PABackbone=BN Inception, Embedding Dimension=5122024.05 | 69.3 | — | — | — | — | — | |
| DiVABackbone=ResNet50, Embedding Dimension=5122024.05 | 69.2 | 79.3 | 87.6 | — | — | — | |
| MS+DASBackbone=ResNet50, Embedding Dimension=5122024.05 | 69.2 | 79.2 | 87.1 | — | — | — | |
| DiVABackbone=ResNet50, Embedding Dimension=5122026.01 | 69.2 | 79.3 | — | — | — | — | |
| MS+DASBackbone=ResNet50, Embedding Dimension=5122026.01 | 69.2 | 79.2 | 87.1 | — | — | — | |
| DRML-PABackbone=BN Inception, Embedding Dimension=5122024.05 | 68.7 | 78.6 | 86.3 | — | — | — | |
| DCML-MDWBackbone=ResNet50, Embedding Dimension=5122024.05 | 68.4 | 77.9 | 86.1 | — | — | — | |
| Proxy AnchorBackbone=BN Inception, Embedding Dimension=5122024.05 | 68.4 | 79.2 | 86.8 | — | — | — | |
| DCML-MDWBackbone=ResNet50, Embedding Dimension=5122026.01 | 68.4 | 77.9 | 86.1 | — | — | — | |
| CPMLBackbone=ResNet50, Embedding Dimension=5122024.05 | 68.3 | 78.7 | 86.2 | — | — | — | |
| HSE-MBackbone=BN Inception, Embedding Dimension=5122024.05 | 67.6 | 78 | 85.8 | — | — | — | |
| MS+DASBackbone=BN Inception, Embedding Dimension=5122024.05 | 67.1 | 78.1 | 86.4 | — | — | — | |
| Group LossEnsemble size=5, Category=Loss + Ensembles2019.12 | 66.9 | 77.1 | 85.4 | 91.5 | — | — | |
| DiVABackbone=BN Inception, Embedding Dimension=5122024.05 | 66.8 | 77.7 | 84.1 | — | — | — | |
| CircleLossBackbone=BN Inception, Embedding Dimension=5122024.05 | 66.7 | 77.4 | 86.2 | — | — | — | |
| ProxyGMLBackbone=BN Inception, Embedding Dimension=5122024.05 | 66.6 | 77.6 | 86.4 | — | — | — | |
| D and CEnsemble size=8, Category=Loss + Ensembles2019.12 | 65.9 | 76.6 | 84.4 | 90.6 | — | — | |
| Group LossEnsemble size=2, Category=Loss + Ensembles2019.12 | 65.8 | 76.7 | 85.2 | 91.2 | — | — | |
| MSEmbedding Size=5122019.04 | 65.7 | 77 | 86.3 | 91.2 | 95 | 97.3 | |
| GPWCategory=Loss + Sampling2019.12 | 65.7 | 77 | 86.3 | 91.2 | — | — | |
| MultiSimilarityBackbone=BN Inception, Embedding Dimension=5122024.05 | 65.7 | 77 | 86.3 | — | — | — | |
| SoftTripleEmbedding Dimension=512, Backbone=Inception2019.09 | 65.4 | 76.4 | 84.5 | 90.4 | — | — | |
| SoftTripleBackbone=BN Inception, Embedding Dimension=5122024.05 | 65.4 | 76.4 | 84.5 | — | — | — | |
| ARCBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%2026.02 | 65.1 | — | — | — | — | — | |
| ESPHNBackbone=ResNet50, Embedding Dimension=5122024.05 | 64.9 | 75.3 | 83.5 | — | — | — | |
| ESPHNBackbone=ResNet50, Embedding Dimension=5122026.01 | 64.9 | 75.3 | 83.5 | — | — | — | |
| BaselineBackbone=SigLip 2 (ViT-B/16), Compression Ratio=0%2026.02 | 64.8 | — | — | — | — | — | |
| Proxy NCA++Backbone=ResNet50, Embedding Dimension=5122024.05 | 64.7 | — | 85.1 | — | — | — | |
| Proxy NCA++Backbone=ResNet50, Embedding Dimension=5122026.01 | 64.7 | — | — | — | — | — | |
| ARCBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%2026.02 | 64.5 | — | — | — | — | — | |
| PTQ-P4QBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%, Bits=8-bit, Fine-tuning=true2026.02 | 64.4 | — | — | — | — | — | |
| SoftMaxnormEmbedding Dimension=5122019.09 | 64.2 | 75.6 | 84.3 | 90.2 | — | — | |
| PTQ-MinMaxBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%, Bits=8-bit, Fine-tuning=true2026.02 | 64.2 | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%2026.02 | 64.1 | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%2026.02 | 63.7 | — | — | — | — | — | |
| MarginEmbedding Size=1282019.04 | 63.6 | 74.4 | 83.1 | 90 | 94.2 | — | |
| MarginEmbedding Dimension=128, Backbone=ResNet-502019.09 | 63.6 | 74.4 | 83.1 | 90 | — | — | |
| Samp. Matt.Category=Loss + Sampling2019.12 | 63.6 | 74.4 | 83.1 | 90 | — | — | |
| ARCBackbone=SigLip 2 (ViT-B/16), Compression Ratio=87.5%2026.02 | 63.4 | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%2026.02 | 62.9 | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%2026.02 | 62.7 | — | — | — | — | — | |
| PTQ-P4QBackbone=SigLip 2 (ViT-B/16), Compression Ratio=87.5%, Bits=4-bit, Fine-tuning=true2026.02 | 61.8 | — | — | — | — | — |