Image Classification on CIFAR-10 (Calibration Evaluation)
1ECEDirichlet
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DirichletBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 1 | 45 | 1 | 0.242 | 12 | 0.2 | 0.5 | |
| Naive CMCEBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation, alpha=0.012025.05 | 1 | 17 | 1 | 0.235 | 11 | 0.2 | 0.6 | |
| IsotonicBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 2 | 34 | 1 | 0.299 | 11 | 0.3 | 0.4 | |
| Temp. scalingBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 2 | 35 | 1 | 0.242 | 11 | 0.3 | 0.3 | |
| V.-Abers (OvA)Backbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 2 | 20 | 1 | 0.246 | 11 | 1 | 0.9 | |
| TSBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation, alpha=0.012025.05 | 2 | 27 | 1 | 0.241 | 11 | 0.3 | 0.1 | |
| Ada-TSBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Adaptive2025.05 | 3 | 35 | 1 | 0.328 | 12 | 0.4 | 0.4 | |
| BaseBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 4 | 32 | 1 | 0.282 | 12 | 0.7 | 0.7 | |
| Platt ScalingBackbone=ResNet-20, Score type=MSP score, Transformation strategy=Constant transformation2025.05 | 4 | 29 | 1 | 0.256 | 12 | 0.6 | 0.8 |