Image Classification on ImageNet (val) (Top-1, Top-5, NLL)
59.15Top-1 AccuracyElliptical-QUEST
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Elliptical-QUESTAttention=Elliptical-QUEST, Attack=SPSA, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 59.15 | 87.44 | 1.613 | |
| EllipticalAttention=Elliptical, Attack=SPSA, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 57.96 | 86.92 | 1.654 | |
| QUESTAttention=QUEST, Attack=SPSA, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 50.7 | 85.49 | 1.904 | |
| StandardAttention=Standard, Attack=SPSA, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 47.17 | 83.95 | 2.027 | |
| Elliptical-QUESTAttention=Elliptical-QUEST, Attack=Auto, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 28.54 | 68.1 | 2.965 | |
| EllipticalAttention=Elliptical, Attack=Auto, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 27.35 | 67.28 | 3.014 | |
| QUESTAttention=QUEST, Attack=Auto, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 27.29 | 67.98 | 3.2 | |
| StandardAttention=Standard, Attack=Auto, Backbone=ViT-Ti/16, Training Recipe=DeiT2026.03 | 26.57 | 67.6 | 3.23 |