Image Classification on NABirds (Top-5 Accuracy)
88.08Top-5 AccuracySCE
Evaluation Results
| Method | Links | |
|---|---|---|
| SCEArchitecture=ResNet-34, Training Protocol=end-to-end training2026.05 | 88.08 | |
| HACEArchitecture=ResNet-34, Training Protocol=end-to-end training2026.05 | 87.67 | |
| HACEArchitecture=ResNet-50, Training Protocol=end-to-end training2026.05 | 87.01 | |
| HACEArchitecture=ResNet-18, Training Protocol=end-to-end training2026.05 | 86.86 | |
| SCEArchitecture=ResNet-18, Training Protocol=end-to-end training2026.05 | 83.73 | |
| SCEArchitecture=ResNet-50, Training Protocol=end-to-end training2026.05 | 80.73 | |
| HACEArchitecture=ConvNeXt-T, Training Protocol=end-to-end training2026.05 | 73.79 | |
| HACEArchitecture=Swin-T, Training Protocol=end-to-end training2026.05 | 69.25 | |
| SCEArchitecture=Swin-T, Training Protocol=end-to-end training2026.05 | 66.16 | |
| SCEArchitecture=ConvNeXt-T, Training Protocol=end-to-end training2026.05 | 53.42 | |
| HACEArchitecture=ViT-B/16, Training Protocol=end-to-end training2026.05 | 53.34 | |
| SCEArchitecture=ViT-B/16, Training Protocol=end-to-end training2026.05 | 51.23 |