Multi-class classification on CIFAR-100 (test)
89.63AccuracyUNCALIBRATED
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| UNCALIBRATEDModel Architecture=ViT2025.12 | 89.63 | 6.36 | 0.448 | 0.16 | 0.0017 | 7.14 | — | — | — | — | |
| TSModel Architecture=ViT2025.12 | 89.63 | 2.96 | 0.396 | 0.122 | 0.0016 | 5.89 | — | — | — | — | |
| VSModel Architecture=ViT2025.12 | 89.59 | 3.67 | 0.439 | 0.137 | 0.0016 | 6.57 | — | — | — | — | |
| NA-FIRModel Architecture=ViT2025.12 | 89.57 | 1.68 | 0.368 | 0.121 | 0.0015 | 4.52 | — | — | — | — | |
| FIRModel Architecture=ViT2025.12 | 89.56 | 2.84 | 0.379 | 0.119 | 0.0015 | 4.8 | — | — | — | — | |
| IR OVRModel Architecture=ViT2025.12 | 89.55 | 3.86 | 1.234 | 0.134 | 0.0016 | 5 | — | — | — | — | |
| SCIRModel Architecture=ViT2025.12 | 89.43 | 1.09 | 0.43 | 0.118 | 0.0015 | 4.39 | — | — | — | — | |
| MSModel Architecture=ViT2025.12 | 88.75 | 5.36 | 0.537 | 0.163 | 0.0018 | 6.62 | — | — | — | — | |
| SCIRModel Architecture=SWIN2025.12 | 87.2 | 0.88 | 0.4698 | 0.13 | 0.0019 | 4.91 | — | — | — | — | |
| UNCALIBRATEDModel Architecture=SWIN2025.12 | 87.19 | 3.28 | 0.4208 | 0.15 | 0.0019 | 5.67 | — | — | — | — | |
| NA-FIRModel Architecture=SWIN2025.12 | 87.19 | 0.84 | 0.4116 | 0.14 | 0.0018 | 4.61 | — | — | — | — | |
| FIRModel Architecture=SWIN2025.12 | 87.19 | 1.33 | 0.4121 | 0.14 | 0.0018 | 5.22 | — | — | — | — | |
| TSModel Architecture=SWIN2025.12 | 87.19 | 1.08 | 0.4112 | 0.14 | 0.0018 | 4.71 | — | — | — | — | |
| VSModel Architecture=SWIN2025.12 | 86.89 | 1.75 | 0.4347 | 0.16 | 0.0019 | 5.59 | — | — | — | — | |
| IR OVRModel Architecture=SWIN2025.12 | 86.49 | 3.3 | 1.3629 | 0.16 | 0.002 | 6.54 | — | — | — | — | |
| MSModel Architecture=SWIN2025.12 | 83.08 | 7.1 | 0.6514 | 0.22 | 0.0025 | 7.86 | — | — | — | — | |
| MixUp Thulasidasan et al. (2019)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 78.3 | — | — | — | — | — | 98.54 | 5.49 | 25 | 0.19 | |
| NA-FIRModel Architecture=ResNet1102025.12 | 77.35 | 5.38 | 0.977 | 0.21 | 0.0033 | 4.48 | — | — | — | — | |
| FIRModel Architecture=ResNet1102025.12 | 77.35 | 7.08 | 1.047 | 0.22 | 0.0034 | 7.14 | — | — | — | — | |
| VSModel Architecture=ResNet1102025.12 | 77.28 | 5.36 | 1.048 | 0.22 | 0.0034 | 7.02 | — | — | — | — | |
| UNCALIBRATEDModel Architecture=ResNet1102025.12 | 77.27 | 19.05 | 1.792 | 0.41 | 0.0041 | 16.61 | — | — | — | — | |
| TSModel Architecture=ResNet1102025.12 | 77.27 | 5.04 | 1.045 | 0.21 | 0.0034 | 7.17 | — | — | — | — | |
| IR OVRModel Architecture=ResNet1102025.12 | 77.11 | 7.36 | 1.539 | 0.21 | 0.0034 | 5.87 | — | — | — | — | |
| SCIRModel Architecture=ResNet1102025.12 | 77 | 1.27 | 0.995 | 0.2 | 0.0032 | 4 | — | — | — | — | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=AdaFocal Ghosh et al. (2022)2026.05 | 76.2 | — | — | — | — | — | 99.1 | 5.4 | 22 | 0.14 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=NLL (CE)2026.05 | 76.11 | — | — | — | — | — | 98.36 | 5.2 | 22 | 0.15 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=CE + TS Guo et al. (2017)2026.05 | 76.11 | — | — | — | — | — | 99.18 | 6.62 | 24 | 0.18 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=CRL Moon et al. (2020)2026.05 | 76.11 | — | — | — | — | — | 98.36 | 5.21 | 22 | 0.12 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=MbLS Liu et al. (2022)2026.05 | 75.93 | — | — | — | — | — | 99.11 | 8.46 | 28 | 0.23 | |
| MbLS Liu et al. (2022)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 75.92 | — | — | — | — | — | 98.46 | 4.8 | 20 | 0.13 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=FL + MDCA Hebbalaguppe et al. (2022b)2026.05 | 75.86 | — | — | — | — | — | 98.29 | 5.25 | 22 | 0.16 | |
| VSModel Architecture=DenseNet1212025.12 | 75.84 | 4.83 | 1.169 | 0.219 | 0.0036 | 6.07 | — | — | — | — | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=LS Szegedy et al. (2015)2026.05 | 75.81 | — | — | — | — | — | 99.13 | 9.15 | 29 | 0.25 | |
| NA-FIRModel Architecture=DenseNet1212025.12 | 75.51 | 4.91 | 1.112 | 0.229 | 0.0036 | 4.12 | — | — | — | — | |
| FIRModel Architecture=DenseNet1212025.12 | 75.51 | 7.56 | 1.194 | 0.238 | 0.0037 | 7.06 | — | — | — | — | |
| UNCALIBRATEDModel Architecture=DenseNet1212025.12 | 75.48 | 20.98 | 2.056 | 0.45 | 0.0045 | 17.08 | — | — | — | — | |
| TSModel Architecture=DenseNet1212025.12 | 75.48 | 5.06 | 1.19 | 0.236 | 0.0037 | 7.14 | — | — | — | — | |
| IR OVRModel Architecture=DenseNet1212025.12 | 75.28 | 7.96 | 1.718 | 0.218 | 0.0037 | 5.27 | — | — | — | — | |
| SCIRModel Architecture=DenseNet1212025.12 | 75.2 | 1.57 | 1.204 | 0.221 | 0.0035 | 3.62 | — | — | — | — | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=MMCE Kumar et al. (2018)2026.05 | 75.04 | — | — | — | — | — | 98.35 | 8.43 | 30 | 0.25 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=MixUp Thulasidasan et al. (2019)2026.05 | 74.81 | — | — | — | — | — | 98.22 | 27.02 | 60 | 0.56 | |
| LS Szegedy et al. (2015)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 74.66 | — | — | — | — | — | 98.3 | 11.02 | 31 | 0.4 | |
| NLL (CE)Refinement Method (Ref.)=true, Calibration Method (Cal.)=false, Backbone=ResNet-502026.05 | 73.89 | — | — | — | — | — | 98.34 | 5.95 | 22 | 0.12 | |
| CE + TS Guo et al. (2017)Refinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 73.89 | — | — | — | — | — | 99.1 | 8 | 28 | 0.23 | |
| CRL Moon et al. (2020)Refinement Method (Ref.)=true, Calibration Method (Cal.)=false, Backbone=ResNet-502026.05 | 73.89 | — | — | — | — | — | 98.29 | 5.94 | 22 | 0.12 | |
| MSModel Architecture=ResNet1102025.12 | 73.88 | 12.91 | 1.843 | 0.34 | 0.0041 | 7.95 | — | — | — | — | |
| FL + MDCA Hebbalaguppe et al. (2022b)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 73.46 | — | — | — | — | — | 98.34 | 5.71 | 22 | 0.14 | |
| MMCE Kumar et al. (2018)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 72.68 | — | — | — | — | — | 98.1 | 8.67 | 26 | 0.13 | |
| RefCalRefinement Method (Ref.)=true, Calibration Method (Cal.)=true, Backbone=ResNet-50, Baseline Comparison=LogitNorm Wei et al. (2022)2026.05 | 72.1 | — | — | — | — | — | 97.55 | 19.63 | 43 | 0.19 | |
| MSModel Architecture=DenseNet1212025.12 | 71.79 | 13.87 | 1.86 | 0.361 | 0.0043 | 8.09 | — | — | — | — | |
| LogitNorm Wei et al. (2022)Refinement Method (Ref.)=true, Calibration Method (Cal.)=false, Backbone=ResNet-502026.05 | 69.89 | — | — | — | — | — | 97.32 | 67.8 | 4 | 1.59 | |
| AdaFocal Ghosh et al. (2022)Refinement Method (Ref.)=false, Calibration Method (Cal.)=true, Backbone=ResNet-502026.05 | 68.49 | — | — | — | — | — | 98.7 | 16.04 | 22 | 0.14 |