Image Classification on TinyImagenet (FLOPs Threshold Accuracy, ECE, EEFP)
49.33Accuracy (25% FLOPs)Early-Exit Distillation
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Early-Exit DistillationBase EENN Framework=Early-Exit Distillation, Calibration Method Variant=Ours, Backbone Architecture=ResNet342025.08 | 49.33 | 62.47 | 65 | 0.07 | 0.78 | |
| Early-Exit DistillationBase EENN Framework=Early-Exit Distillation, Calibration Method Variant=Calibrated, Backbone Architecture=ResNet342025.08 | 48.11 | 61.7 | 64.21 | 0.02 | 0.74 | |
| Zero Time WasteBase EENN Framework=Zero Time Waste, Calibration Method Variant=Ours, Backbone Architecture=ResNet342025.08 | 48 | 60.03 | 63.24 | 0.08 | 0.75 | |
| Gradient EquilibriumBase EENN Framework=Gradient Equilibrium, Calibration Method Variant=Ours, Backbone Architecture=ResNet342025.08 | 47.7 | 61.31 | 64.87 | 0.07 | 0.77 | |
| Early-Exit DistillationBase EENN Framework=Early-Exit Distillation, Calibration Method Variant=Baseline, Backbone Architecture=ResNet342025.08 | 47.35 | 61.15 | 63.84 | 0.13 | 0.73 | |
| Gradient EquilibriumBase EENN Framework=Gradient Equilibrium, Calibration Method Variant=Calibrated, Backbone Architecture=ResNet342025.08 | 46.39 | 60.33 | 64.28 | 0.02 | 0.73 | |
| Zero Time WasteBase EENN Framework=Zero Time Waste, Calibration Method Variant=Calibrated, Backbone Architecture=ResNet342025.08 | 46.34 | 59.45 | 62.38 | 0.01 | 0.72 | |
| Gradient EquilibriumBase EENN Framework=Gradient Equilibrium, Calibration Method Variant=Baseline, Backbone Architecture=ResNet342025.08 | 46.07 | 59.59 | 64.01 | 0.12 | 0.72 | |
| Zero Time WasteBase EENN Framework=Zero Time Waste, Calibration Method Variant=Baseline, Backbone Architecture=ResNet342025.08 | 45.8 | 59.04 | 62.22 | 0.13 | 0.71 |