Deep Metric Learning on Stanford Online Products (SOP)
81.4R@1IB-DML
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| IB-DMLBackbone=ResNet50, Embedding Dimension=512, Optimizer=RAdam, Input Image Size=Larger, Batch Size=Larger on SOP2022.03 | 81.4 | 91.3 | 92.6 | |
| S2SD+PLGBackbone=ResNet50, Embedding Dimension=5122022.03 | 81.3 | 92.3 | 91.1 | |
| S2SD+PLGBackbone=ResNet50, Embedding Dimension=1282022.03 | 80.6 | 91.8 | 90.9 | |
| S2SDBackbone=ResNet50, Embedding Dimension=1282022.03 | 80.2 | 91.5 | 90.9 | |
| DCML-MDWBackbone=ResNet50, Embedding Dimension=5122022.03 | 79.8 | 90.8 | 90.8 | |
| DiVABackbone=ResNet50, Embedding Dimension=5122022.03 | 79.6 | 91.2 | 90.6 | |
| S2SD+PLGBackbone=Inception-BN, Embedding Dimension=5122022.03 | 79.4 | 91.2 | 90.4 | |
| ProxyAnchorBackbone=Inception-BN, Embedding Dimension=512, Pooling Operations=Combination in backbone2022.03 | 79.1 | 90.8 | — | |
| Multisimilarity+PLGBackbone=ResNet50, Embedding Dimension=5122022.03 | 79 | 91 | 90 | |
| Multisimilarity+PLGBackbone=Inception-BN, Embedding Dimension=5122022.03 | 78.6 | 90.7 | 90 | |
| EPSHNBackbone=ResNet50, Embedding Dimension=5122022.03 | 78.3 | 90.7 | — | |
| NormSoftBackbone=ResNet50, Embedding Dimension=5122022.03 | 78.2 | 90.6 | — | |
| MultisimilarityBackbone=Inception-BN, Embedding Dimension=5122022.03 | 78.2 | 90.5 | — | |
| ProxyGMLBackbone=Inception-BN, Embedding Dimension=5122022.03 | 78 | 90.6 | 90.2 | |
| Multisimilarity+PLGBackbone=ResNet50, Embedding Dimension=1282022.03 | 77.9 | 89.9 | 90.2 | |
| MICBackbone=ResNet50, Embedding Dimension=1282022.03 | 77.2 | 89.4 | 90 | |
| PADSBackbone=ResNet50, Embedding Dimension=1282022.03 | 76.5 | 89 | 89.9 | |
| Div&ConqBackbone=ResNet50, Embedding Dimension=1282022.03 | 75.9 | 88.4 | 90.2 | |
| GroupBackbone=Inception-BN, Embedding Dimension=5122022.03 | 75.1 | 87.5 | 90.8 | |
| MarginBackbone=ResNet50, Embedding Dimension=1282022.03 | 72.7 | 86.2 | 90.7 |