Image Classification on ImageNet FGSM
56.9Top-1 AccuracyQUEST
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| QUESTBackbone=ViT-Ti/16, Training Recipe=DeiT, Attention mechanism=QUEST2026.03 | 56.9 | — | 86.63 | 1.745 | |
| Elliptical-QUESTBackbone=ViT-Ti/16, Training Recipe=DeiT, Attention mechanism=Elliptical-QUEST2026.03 | 56.39 | — | 85.94 | 1.741 | |
| EllipticalBackbone=ViT-Ti/16, Training Recipe=DeiT, Attention mechanism=Elliptical2026.03 | 55.96 | — | 85.53 | 1.746 | |
| Swin-ACMoE-Top 2Top-k routing=22025.02 | 55.78 | — | 85.8 | — | |
| Swin-Top 2Top-k routing=22025.02 | 54.7 | — | 85.22 | — | |
| StandardBackbone=ViT-Ti/16, Training Recipe=DeiT, Attention mechanism=Standard2026.03 | 54.23 | — | 85.28 | 1.827 | |
| Swin-ACMoE-Top 1Top-k routing=12025.02 | 53.43 | — | 82.8 | — | |
| Swin-Top 1Top-k routing=12025.02 | 52.84 | — | 83.86 | — | |
| PiT-S2021.03 | — | 29.5 | — | — | |
| ResNet502021.03 | — | 7.1 | — | — | |
| ResNet50variant=well-optimized (†)2021.03 | — | 24.7 | — | — | |
| ViT-S2021.03 | — | 27.2 | — | — |