Image Classification on ImageNet-1k (Speed-up and Delta Accuracy)
3.31Speed-upSPoT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SPoTPre-training=CLS-IN21k, Token Budget=100, Reduction Stage=inference2025.07 | 3.31 | -2.95 | |
| SPoTPre-training=MAE-IN1k, Token Budget=100, Reduction Stage=inference2025.07 | 3.31 | -1.13 | |
| ToMePre-training=CLS-IN21k, Token Budget=100, Reduction Stage=inference2025.07 | 1.95 | -4.2 | |
| ToMePre-training=MAE-IN1k, Token Budget=100, Reduction Stage=finetuning2025.07 | 1.95 | -1.71 | |
| ToMePre-training=MAE-IN1k, Token Budget=100, Reduction Stage=inference2025.07 | 1.94 | -4.87 |