Image Classification on ImageNet ILSVRC2012 (val) (Accuracy and Efficiency)
81.8Top-1 AccuracyBaseline
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BaselineModel=ViT (DeiT)-B2021.06 | 81.8 | 95.6 | 17.6 | 0 | 292 | 0 | — | |
| BaselineBackbone=DeiT-B2023.10 | 81.8 | — | 17.6 | — | 295 | — | 86.6 | |
| DPS-ViTModel=ViT (DeiT)-B2021.06 | 81.6 | 95.4 | 9.4 | 46.6 | 413 | 41.3 | — | |
| PS-ViTModel=ViT (DeiT)-B2021.06 | 81.5 | 95.4 | 9.8 | 44.3 | 414 | 41.8 | — | |
| BaselineModel=T2T-ViT-142021.06 | 81.5 | 95.4 | 5.2 | 0 | 764 | 0 | — | |
| PPTBackbone=DeiT-B, Protocol=fine-tuned2023.10 | 81.4 | — | 11.6 | — | 445 | — | 86.6 | |
| DPS-T2TModel=T2T-ViT-142021.06 | 81.3 | 95.3 | 3.1 | 45.4 | 1,078 | 41.1 | — | |
| DynamicViTBackbone=DeiT-B2023.10 | 81.3 | — | 11.5 | — | 454 | — | 89.4 | |
| Evo-ViTBackbone=DeiT-B, Training=scratch (300 epochs)2023.10 | 81.3 | — | 11.7 | — | 429 | — | 87.3 | |
| EVITBackbone=DeiT-B2023.10 | 81.3 | — | 11.6 | — | 440 | — | 86.6 | |
| PS-T2TModel=T2T-ViT-142021.06 | 81.1 | 95.2 | 3.1 | 40.4 | 1,055 | 38.1 | — | |
| VTPModel=ViT (DeiT)-B2021.06 | 80.7 | 95 | 10 | 43.2 | 412 | 41 | — | |
| IA-RED2Backbone=DeiT-B2023.10 | 80.3 | — | 11.8 | — | 453 | — | — | |
| PPTBackbone=DeiT-B, Protocol=off-the-shelf2023.10 | 80.3 | — | 11.6 | — | 445 | — | 86.6 | |
| POWERModel=ViT (DeiT)-B2021.06 | 80.1 | 94.6 | 10.4 | 39.2 | 397 | 35.8 | — | |
| POWERModel=T2T-ViT-142021.06 | 79.9 | 94.4 | 3.5 | 32.7 | 991 | 29.7 | — | |
| BaselineModel=ViT (DeiT)-S2021.06 | 79.8 | 95 | 4.6 | 0 | 940 | 0 | — | |
| BaselineBackbone=DeiT-S2023.10 | 79.8 | — | 4.6 | — | 993 | — | 22.1 | |
| PPTBackbone=DeiT-S, Protocol=fine-tuned2023.10 | 79.8 | — | 2.9 | — | 1,448 | — | 22.1 | |
| SCOPModel=ViT (DeiT)-B2021.06 | 79.7 | 94.5 | 10.2 | 42 | 403 | 38.1 | — | |
| ATSBackbone=DeiT-S2023.10 | 79.7 | — | 2.9 | — | 1,382 | — | 22.1 | |
| TPSBackbone=DeiT-S2023.10 | 79.7 | — | 3 | — | — | — | 22.1 | |
| DPS-ViTModel=ViT (DeiT)-S2021.06 | 79.5 | 94.8 | 2.4 | 47.8 | 1,342 | 42.8 | — | |
| EVITBackbone=DeiT-S2023.10 | 79.5 | — | 3 | — | 1,378 | — | 22.1 | |
| PPTBackbone=DeiT-S, Protocol=off-the-shelf2023.10 | 79.5 | — | 2.9 | — | 1,448 | — | 22.1 | |
| PS-ViTModel=ViT (DeiT)-S2021.06 | 79.4 | 94.7 | 2.6 | 43.6 | 1,321 | 40.5 | — | |
| PS-ViTBackbone=DeiT-S2023.10 | 79.4 | — | 2.6 | — | 1,321 | — | — | |
| Evo-ViTBackbone=DeiT-S, Training=scratch (300 epochs)2023.10 | 79.4 | — | 3 | — | 1,414 | — | 22.4 | |
| ToMeBackbone=DeiT-S2023.10 | 79.4 | — | 2.7 | — | 1,552 | — | 22.1 | |
| DynamicViTBackbone=DeiT-S2023.10 | 79.3 | — | 3 | — | 1,440 | — | 22.8 | |
| IA-RED2Backbone=DeiT-S2023.10 | 79.1 | — | 3.2 | — | 1,362 | — | — | |
| POWERModel=ViT (DeiT)-S2021.06 | 78.3 | 94 | 2.7 | 41.3 | 1,295 | 37.8 | — | |
| HVTModel=ViT (DeiT)-S2021.06 | 78 | 93.8 | 2.4 | 47.8 | 1,335 | 42.1 | — | |
| SCOPModel=ViT (DeiT)-S2021.06 | 77.5 | 93.5 | 2.6 | 43.6 | 1,310 | 39.4 | — | |
| BaselineModel=ViT (DeiT)-Ti2021.06 | 72.2 | 91.1 | 1.3 | 0 | 2,536 | 0 | — | |
| BaselineBackbone=DeiT-Ti2023.10 | 72.2 | — | 1.3 | — | 2,675 | — | 5.6 | |
| DPS-ViTModel=ViT (DeiT)-Ti2021.06 | 72.1 | 91.1 | 0.6 | 53.8 | 3,639 | 43.5 | — | |
| PPTBackbone=DeiT-Ti, Protocol=fine-tuned2023.10 | 72.1 | — | 0.8 | — | 3,572 | — | 5.6 | |
| PS-ViTModel=ViT (DeiT)-Ti2021.06 | 72 | 91 | 0.7 | 46.2 | 3,576 | 41 | — | |
| Evo-ViTBackbone=DeiT-Ti, Training=scratch (300 epochs)2023.10 | 72 | — | 0.8 | — | 3,781 | — | 5.9 | |
| EVITBackbone=DeiT-Ti2023.10 | 71.9 | — | 0.8 | — | 3,387 | — | 5.6 | |
| PPTBackbone=DeiT-Ti, Protocol=off-the-shelf2023.10 | 71.6 | — | 0.8 | — | 3,572 | — | 5.6 | |
| DynamicViTBackbone=DeiT-Ti2023.10 | 71.4 | — | 0.8 | — | 3,765 | — | 5.9 | |
| ToMeBackbone=DeiT-Ti2023.10 | 71.4 | — | 0.8 | — | 3,685 | — | 5.6 | |
| HVTModel=ViT (DeiT)-Ti2021.06 | 69.7 | 89.4 | 0.7 | 46.2 | 3,524 | 38.9 | — | |
| POWERModel=ViT (DeiT)-Ti2021.06 | 69.4 | 89.2 | 0.8 | 38.4 | 3,304 | 30.3 | — | |
| SCOPModel=ViT (DeiT)-Ti2021.06 | 68.9 | 89 | 0.8 | 38.4 | 3,372 | 33 | — |