Object Instance Retrieval on Stanford Online Products (test)
84.74R@1RRT + R50 w/ LEPI
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RRT + R50 w/ LEPITraining status=Trained on SOP, Backbone=ResNet-50, Positional Encoding type=LEPI2022.11 | 84.74 | 93.29 | 96.04 | 37.25 | |
| RRT + R50 w/ LMaxEPITraining status=Trained on SOP, Backbone=ResNet-50, Positional Encoding type=LMaxEPI2022.11 | 84.53 | 93.27 | 96.04 | 37.19 | |
| RRT + R50 (finetune)Training status=Trained on SOP, Backbone=ResNet-50, Training protocol=finetune2022.11 | 84.46 | 93.21 | 96.04 | 37.14 | |
| RRT + R50 w/ EPETraining status=Trained on SOP, Backbone=ResNet-50, Positional Encoding type=EPE2022.11 | 82.57 | 92.69 | 95.89 | 35.38 | |
| RRT + R50 (frozen)Training status=Trained on SOP, Backbone=ResNet-50, Training protocol=frozen2022.11 | 81.8 | 92.35 | 95.78 | 34.91 | |
| Cross-Entropy basedTraining status=Trained on SOP2022.11 | 81.1 | 91.7 | — | — | |
| R50 (trained)Training status=Trained on SOP, Backbone=ResNet-502022.11 | 80.74 | 91.87 | 95.54 | 32.9 | |
| XBMTraining status=Trained on SOP2022.11 | 80.6 | 91.6 | — | — | |
| Margin-basedTraining status=Trained on SOP2022.11 | 76.1 | 88.4 | — | — | |
| FastAPTraining status=Trained on SOP2022.11 | 73.8 | 88 | — | — | |
| R50Training status=Pretrained, Backbone=ResNet-502022.11 | 55.89 | 71.32 | 79.69 | 12.09 | |
| VGG16Training status=Pretrained, Backbone=VGG162022.11 | 55.75 | 70.86 | 79.65 | 11.93 | |
| NetVLADTraining status=Pretrained, Backbone=NetVLAD2022.11 | 54.16 | 70.85 | 79.62 | 11.9 |