Image Classification on ImageNet 1k (val) (Exit Layer Improvement Metrics)
55.16Rel Imp RG(ŷ)(0-1) ε=0.01Dyn-Perc
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Dyn-PercRisk control method=CRC, Calibration set size (n)=1000, Backbone architecture=Dyn-Perc2024.05 | 55.16 | 66.73 | 54.28 | 65.25 | 43.35 | 58.62 | 53.67 | 62.04 | |
| DVITRisk control method=CRC, Calibration set size (n)=1000, Backbone architecture=DVIT2024.05 | 48.68 | 61.6 | 39.05 | 54.89 | 40.75 | 52.21 | 38.48 | 49.86 | |
| L2W-DENRisk control method=CRC, Calibration set size (n)=1000, Backbone architecture=L2W-DEN2024.05 | 47.73 | 65.04 | 35.09 | 53.81 | 37.08 | 51.7 | 33.76 | 48.48 | |
| MSDNetRisk control method=CRC, Calibration set size (n)=1000, Backbone architecture=MSDNet2024.05 | 46.56 | 62.55 | 33.11 | 50.56 | 33.01 | 46.12 | 34.09 | 46.26 |