Image Classification on ImageNet-LT ILSVRC2012 (test)
35.6Top-1 Accuracy (Overall)OLTR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OLTRBackbone=ResNet-10, Training Setting=From scratch2020.08 | 35.6 | 43.2 | 35.1 | 18.5 | |
| Feature Space AugmentationBackbone=ResNet-10, Training Setting=From scratch, Loss=Focal Loss2020.08 | 35.3 | 47 | 31.3 | 16.8 | |
| Feature Space AugmentationBackbone=ResNet-10, Training Setting=From scratch, Loss=Cross-Entropy2020.08 | 35.2 | 47.3 | 31.6 | 14.7 | |
| Lifted LossBackbone=ResNet-10, Training Setting=From scratch2020.08 | 30.8 | 35.8 | 30.4 | 17.9 | |
| Range LossBackbone=ResNet-10, Training Setting=From scratch2020.08 | 30.7 | 35.8 | 30.3 | 17.6 | |
| FLBackbone=ResNet-10, Training Setting=From scratch2020.08 | 30.5 | 36.4 | 29.9 | 16 | |
| FSLWFBackbone=ResNet-10, Training Setting=From scratch2020.08 | 28.4 | 40.9 | 22.1 | 15 | |
| Plain ModelBackbone=ResNet-10, Training Setting=From scratch2020.08 | 20.9 | 40.9 | 10.7 | 0.4 |