Multi-Class Calibration on CIFAR-100 (test) logits
-0.97LogLoss Absolute ImprovementSMS
Evaluation results
| Method | Links | |
|---|---|---|
| SMSBackbone=densenet402025.11 | -0.97 | |
| Structured Matrix Scaling (SMS)Model=densenet402025.11 | -0.97 | |
| SVSBackbone=densenet402025.11 | -0.967 | |
| Structured Vector Scaling (SVS)Model=densenet402025.11 | -0.967 | |
| TS probmetricsBackbone=densenet402025.11 | -0.961 | |
| Temperature Scaling (probmetrics)Model=densenet402025.11 | -0.961 | |
| VS torchcalBackbone=densenet402025.11 | -0.958 | |
| Vector Scaling (torchcal)Model=densenet402025.11 | -0.958 | |
| SMSBackbone=resnet-wide322025.11 | -0.869 | |
| Structured Matrix Scaling (SMS)Model=resnet-wide322025.11 | -0.869 | |
| SVSBackbone=resnet-wide322025.11 | -0.864 | |
| Structured Vector Scaling (SVS)Model=resnet-wide322025.11 | -0.864 | |
| TS probmetricsBackbone=resnet-wide322025.11 | -0.858 | |
| Temperature Scaling (probmetrics)Model=resnet-wide322025.11 | -0.858 | |
| VS torchcalBackbone=resnet-wide322025.11 | -0.851 | |
| Vector Scaling (torchcal)Model=resnet-wide322025.11 | -0.851 | |
| SVSBackbone=resnet1102025.11 | -0.608 | |
| SMSBackbone=resnet1102025.11 | -0.608 | |
| Structured Vector Scaling (SVS)Model=resnet1102025.11 | -0.608 | |
| Structured Matrix Scaling (SMS)Model=resnet1102025.11 | -0.608 | |
| TS probmetricsBackbone=resnet1102025.11 | -0.602 | |
| Temperature Scaling (probmetrics)Model=resnet1102025.11 | -0.602 | |
| VS torchcalBackbone=resnet1102025.11 | -0.598 | |
| Vector Scaling (torchcal)Model=resnet1102025.11 | -0.598 | |
| SMSBackbone=resnet110-SD2025.11 | -0.437 | |
| Structured Matrix Scaling (SMS)Model=resnet110-SD2025.11 | -0.437 | |
| SVSBackbone=resnet110-SD2025.11 | -0.423 | |
| Structured Vector Scaling (SVS)Model=resnet110-SD2025.11 | -0.423 | |
| VS torchcalBackbone=resnet110-SD2025.11 | -0.421 | |
| Vector Scaling (torchcal)Model=resnet110-SD2025.11 | -0.421 | |
| TS probmetricsBackbone=resnet110-SD2025.11 | -0.41 | |
| Temperature Scaling (probmetrics)Model=resnet110-SD2025.11 | -0.41 | |
| SMSBackbone=lenet52025.11 | -0.273 | |
| Structured Matrix Scaling (SMS)Model=lenet52025.11 | -0.273 | |
| VS torchcalBackbone=lenet52025.11 | -0.266 | |
| Vector Scaling (torchcal)Model=lenet52025.11 | -0.266 | |
| SVSBackbone=lenet52025.11 | -0.219 | |
| Structured Vector Scaling (SVS)Model=lenet52025.11 | -0.219 | |
| TS probmetricsBackbone=lenet52025.11 | -0.134 | |
| Temperature Scaling (probmetrics)Model=lenet52025.11 | -0.134 | |
| DirichletBackbone=lenet52025.11 | 0.166 | |
| Dirichlet CalibrationModel=lenet52025.11 | 0.166 | |
| DirichletBackbone=resnet1102025.11 | 0.4 | |
| Dirichlet CalibrationModel=resnet1102025.11 | 0.4 | |
| DirichletBackbone=resnet-wide322025.11 | 0.54 | |
| Dirichlet CalibrationModel=resnet-wide322025.11 | 0.54 | |
| DirichletBackbone=densenet402025.11 | 1.075 | |
| Dirichlet CalibrationModel=densenet402025.11 | 1.075 | |
| DirichletBackbone=resnet110-SD2025.11 | 1.201 | |
| Dirichlet CalibrationModel=resnet110-SD2025.11 | 1.201 | |
| MS torchcalBackbone=lenet52025.11 | 8.204 | |
| Matrix Scaling (torchcal)Model=lenet52025.11 | 8.204 | |
| MS torchcalBackbone=resnet-wide322025.11 | 25.516 | |
| Matrix Scaling (torchcal)Model=resnet-wide322025.11 | 25.516 | |
| MS torchcalBackbone=densenet402025.11 | 28.982 | |
| Matrix Scaling (torchcal)Model=densenet402025.11 | 28.982 | |
| MS torchcalBackbone=resnet1102025.11 | 35.361 | |
| Matrix Scaling (torchcal)Model=resnet1102025.11 | 35.361 | |
| MS torchcalBackbone=resnet110-SD2025.11 | 37.586 | |
| Matrix Scaling (torchcal)Model=resnet110-SD2025.11 | 37.586 |