Image Classification on ImageNet (val) (Top-1, Peak, Final)
83.5Top-1 AccuracySwin-B (Baseline)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Swin-B (Baseline)Params=88M, Mod.=-2025.11 | 83.5 | 4,824 | 876 | |
| Swin-B TWEOParams=88M, Mod.=No2025.11 | 83.4 | 23 | 8 | |
| Swin-S TWEOParams=50M, Mod.=No2025.11 | 82.8 | 22 | 10 | |
| Swin-S (Baseline)Params=50M, Mod.=-2025.11 | 82.7 | 6,402 | 1,758 | |
| Swin-T softmax+1Params=28M, Mod.=Yes2025.11 | 81.4 | 811 | 143 | |
| Swin-T TWEOParams=28M, Mod.=No2025.11 | 81.4 | 22 | 15 | |
| ViT-B (Baseline)Params=87M, Mod.=-2025.11 | 81.3 | 1,579 | 106 | |
| ViT-B TWEOParams=87M, Mod.=No2025.11 | 81.3 | 38 | 16 | |
| Swin-T (Baseline)Params=28M, Mod.=-2025.11 | 81.2 | 1,556 | 534 | |
| Swin-T attn biasParams=28M, Mod.=Yes2025.11 | 81.1 | 478 | 135 | |
| Swin-T gatedParams=28M, Mod.=Yes2025.11 | 80.7 | 976 | 126 | |
| ViT-S (Baseline)Params=22M, Mod.=-2025.11 | 79.8 | 328 | 174 | |
| ViT-S TWEOParams=22M, Mod.=No2025.11 | 79.6 | 36 | 19 |