Image Classification on ImageNet-1k (val) (T1/T5 Threshold Accuracy and AuC)
83.11Top-1 Accuracy @ 0% ThresholdCViT
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CViTTraining Mask=Collapse order2025.11 | 83.11 | 96.5 | 81.37 | 95.69 | 79.39 | 94.63 | 70.57 | 88.94 | 72.19 | |
| RViTTraining Mask=Random2025.11 | 83.1 | 96.46 | 81.33 | 95.6 | 78.94 | 94.5 | 67.27 | 87.23 | 70.86 | |
| ViTMasking=Random during inference2025.11 | 82.91 | 96.28 | 80.03 | 94.8 | 74.38 | 91.48 | 22.38 | 36.76 | 57.16 | |
| ViT+CMasking=Collapse masks, Training=Randomly masked2025.11 | 82.84 | 96.23 | 78.09 | 93.82 | 71.67 | 89.58 | 31.04 | 49.45 | 57.27 | |
| DynamicViTBackbone=DeiT-B, Type=Token-pruning2025.11 | 81.74 | 95.64 | 81.44 | 95.46 | 77.54 | 93.3 | 20.66 | 37.09 | 56.32 |