Deep Metric Learning on CARS196
97.4Recall@1VPTSP (CLIP vision)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VPTSP (CLIP vision)Architecture=Vit-L/14, Pre-training Set=Laion2b2024.02 | 97.4 | — | — | — | |
| VPTSP-GBackbone=ViT-B16, Feature Dimension=512, Batch Size=64/322024.02 | 91.2 | 95.1 | — | 97.3 | |
| S2SD+PLGBackbone=ResNet50, Embedding Dimension=5122022.03 | 90.2 | 94.4 | 72.4 | — | |
| RS@KBackbone=ViT-B16, Feature Dimension=512, Batch Size=>2002024.02 | 89.5 | 94.2 | — | 96.6 | |
| S2SD+PLGBackbone=ResNet50, Embedding Dimension=1282022.03 | 89.1 | 93.8 | 73.1 | — | |
| IB-DMLBackbone=ResNet50, Embedding Dimension=512, Optimizer=RAdam, Input Image Size=Larger2022.03 | 88.1 | 93.3 | 74.8 | — | |
| S2SD+PLGBackbone=Inception-BN, Embedding Dimension=5122022.03 | 88.1 | 92.9 | 71.4 | — | |
| VPTSP-GBackbone=ViT-S16, Feature Dimension=384, Batch Size=64/322024.02 | 87.7 | 93.3 | — | 96.1 | |
| DiVA-R50-512Dim=512, Backbone=ResNet502020.04 | 87.6 | 92.9 | 72.2 | — | |
| S2SDBackbone=ResNet50, Embedding Dimension=1282022.03 | 87.6 | 92.7 | 72.3 | — | |
| DiVABackbone=ResNet50, Embedding Dimension=5122022.03 | 87.6 | 92.9 | 72.2 | — | |
| Trip-DiVA-R50-512Dim=512, Backbone=ResNet502020.04 | 87.3 | 92.8 | 72.1 | — | |
| Multisimilarity+PLGBackbone=ResNet50, Embedding Dimension=5122022.03 | 87.1 | 92.3 | 73 | — | |
| VPTSP (DINO)Architecture=Vit-S/16, Pre-training Set=ImageNet 1K2024.02 | 86.9 | — | — | — | |
| ProxyAnchorBackbone=Inception-BN, Embedding Dimension=512, Pooling Operations=Combination in backbone2022.03 | 86.8 | 91.6 | — | — | |
| VPTSP-MBackbone=ViT-S16, Feature Dimension=384, Batch Size=64/322024.02 | 86.8 | 92 | — | 95.5 | |
| Hyp-ViTBackbone=ViT-S16, Feature Dimension=384, Batch Size=882/9002024.02 | 86.5 | 92.1 | — | 95.3 | |
| VPT-BaseBackbone=ViT-S16, Feature Dimension=384, Batch Size=64/322024.02 | 86.3 | 91.8 | — | 95.5 | |
| Multisimilarity+PLGBackbone=Inception-BN, Embedding Dimension=5122022.03 | 86.2 | 91.5 | 70.8 | — | |
| DREMLDim=9216, Type=Ensemble2020.04 | 86 | 91.7 | 76.4 | — | |
| Multisimilarity+PLGBackbone=ResNet50, Embedding Dimension=1282022.03 | 86 | 91.4 | 72.4 | — | |
| GroupBackbone=Inception-BN, Embedding Dimension=5122022.03 | 85.6 | 91.2 | 72.7 | — | |
| ProxyGMLBackbone=Inception-BN, Embedding Dimension=5122022.03 | 85.5 | 91.8 | 72.4 | — | |
| ABEDim=512, Type=Ensemble2020.04 | 85.2 | 90.5 | — | — | |
| DCML-MDWBackbone=ResNet50, Embedding Dimension=5122022.03 | 85.2 | 91.8 | 73.9 | — | |
| DR-MSBackbone=Inception-BN, Embedding Dimension=5122022.03 | 85 | 90.5 | — | — | |
| D&CDim=1282020.04 | 84.6 | 90.7 | 70.3 | — | |
| Div&ConqBackbone=ResNet50, Embedding Dimension=1282022.03 | 84.6 | 90.7 | 70.3 | — | |
| SofttripleDim=5122020.04 | 84.5 | 90.7 | 70.1 | — | |
| VPTSP (DeiT)Architecture=Vit-S/16, Pre-training Set=ImageNet 1K2024.02 | 84.5 | — | — | — | |
| NormSoftBackbone=ResNet50, Embedding Dimension=5122022.03 | 84.2 | 90.4 | — | — | |
| MultiSimDim=5122020.04 | 84.1 | 90.4 | — | — | |
| DiVA-IBN-512Dim=512, Backbone=Inception-BN2020.04 | 84.1 | 90.7 | 68.7 | — | |
| MultisimilarityBackbone=Inception-BN, Embedding Dimension=5122022.03 | 84.1 | 90.4 | — | — | |
| HORDE + Contr.Dim=5122020.04 | 83.9 | 90.3 | — | — | |
| PADSBackbone=ResNet50, Embedding Dimension=1282022.03 | 83.5 | 89.7 | 68.8 | — | |
| VPTSP (MAE)Architecture=Vit-B/16, Pre-training Set=ImageNet 1K2024.02 | 83.2 | — | — | — | |
| Trip-DiVA-IBN-512Dim=512, Backbone=Inception-BN2020.04 | 83.1 | 90.3 | 68.8 | — | |
| EPSHNBackbone=ResNet50, Embedding Dimension=5122022.03 | 82.7 | 89.3 | — | — | |
| MICDim=1282020.04 | 82.6 | 89.1 | 68.4 | — | |
| MICBackbone=ResNet50, Embedding Dimension=1282022.03 | 82.6 | 89.1 | 68.4 | — | |
| Margin-R50-512Dim=512, Backbone=ResNet502020.04 | 82.2 | 89 | 68.1 | — | |
| RankDim=1536, Type=Ensemble2020.04 | 82.1 | 89.3 | 71.8 | — | |
| DVMLDim=5122020.04 | 82 | 88.4 | 67.6 | — | |
| A-BIERDim=512, Type=Ensemble2020.04 | 82 | 89 | — | — | |
| HTLDim=5122020.04 | 81.4 | 88 | — | — | |
| MarginDim=1282020.04 | 79.6 | 86.5 | 69.1 | — | |
| MarginBackbone=ResNet50, Embedding Dimension=1282022.03 | 79.6 | 86.5 | 69.1 | — | |
| HDMLDim=5122020.04 | 79.1 | 87.1 | 69.7 | — | |
| HTGDim=5122020.04 | 76.5 | 84.7 | — | — |