List your company

Model Calibration on Tiny ImageNet

0.82Expected Calibration ErrorBrier Loss
-0.36167.614215.5923.5658Apr 14, 2025Jun 22, 2025Aug 30, 2025Nov 7, 2025Jan 15, 2026Mar 25, 2026Jun 3, 2026

Evaluation results

MethodLinks
Brier LossModel=ResNet-502026.040.82—————3.89——
AFLModel=ResNet-502026.040.83—————2.69——
FLSDModel=ResNet-502026.040.94—————1.82——
AFLModel=ResNet-1102026.040.97—————2.74——
DFLModel=ResNet-502026.041.02—————2.66——
FLSDModel=ResNet-1102026.041.03—————2.22——
DFLModel=ResNet-1102026.041.04—————2.55——
GCEModel=ResNet-1102026.041.15—————4.2——
GCEModel=ResNet-502026.041.19—————4.29——
BalCALBackbone=ResNet-1012025.041.35———66.91.48———
FLSDBackbone=ResNet-1012025.041.39———63.091.65———
BalCALBackbone=ResNet-502025.041.4———65.221.47———
MITBackbone=ResNet-502025.041.45———64.271.32———
Brier LossModel=ResNet-1102026.041.5—————4.15——
MITBackbone=ResNet-1012025.041.58———66.811.64———
LSBackbone=ResNet-502025.041.59———63.11.71———
CEModel=ResNet-502026.041.59—————16.52——
MMCEModel=ResNet-1102026.041.61—————16.25——
CEModel=ResNet-1102026.041.62—————16.07——
MMCEModel=ResNet-502026.041.66—————14.88——
FLSDBackbone=ResNet-502025.041.84———63.091.84———
LSBackbone=ResNet-1012025.041.93———63.126.25———
FLBackbone=ResNet-502025.042.07———62.822.05———
AlignmixupBackbone=ResNet18, Training=from scratch2026.062.18——————6.131.98
OA-CutMix/absBackbone=DeiT-Ti, latency [ms]=1.012026.062.31——————3.27—
VanillaBackbone=ResNet-1012025.042.5———62.052.44———
FLBackbone=ResNet-1012025.042.57———62.377.76———
OA-CutMix/abs + decoupledBackbone=DeiT-Ti, latency [ms]=1.012026.062.84——————3.66—
VTSTBackbone=ResNet-1012025.042.91———60.852.88———
OA-CutMix/absBackbone=ResNet18, Training=from scratch2026.063.11——————4.271.05
OA-CutMix/abs + decoupledBackbone=ResNet18, Training=from scratch2026.063.14——————4.851.01
OA-CutMix/rel + decoupledBackbone=ResNet18, Training=from scratch2026.063.18——————4.661
Mixup+CutMixBackbone=ResNet50, Training=from scratch2026.063.44——————12.250.67
OA-CutMix/relBackbone=ResNet18, Training=from scratch2026.063.49——————4.761.02
OA-CutMix/relBackbone=DeiT-Ti, latency [ms]=1.032026.063.6——————3.96—
PuzzleMixBackbone=DeiT-S, latency [ms]=273.282026.063.81——————6.1—
mixupBackbone=ResNet-502025.043.85———65.883.84———
CutMix + decoupledBackbone=DeiT-Ti, latency [ms]=0.672026.063.88——————4.71—
mixupBackbone=ResNet-1012025.043.98———66.563.86———
OA-CutMix/rel + decoupledBackbone=DeiT-Ti, latency [ms]=1.112026.063.99——————4.6—
MixupBackbone=DeiT-Ti, latency [ms]=0.732026.064.02——————6.03—
Mixup + CutMixBackbone=DeiT-Ti, latency [ms]=0.702026.064.08——————7.41—
PuzzleMixBackbone=DeiT-Ti, latency [ms]=273.282026.064.13——————6.1—
OA-CutMix/absBackbone=ResNet50, Training=from scratch2026.064.19——————4.271.05
VanillaBackbone=ResNet-502025.044.38———60.534.37———
TokenMixBackbone=DeiT-S, latency [ms]=39.912026.064.45——————7.52—
SnapMixBackbone=DeiT-Ti, latency [ms]=39.982026.064.69——————10.15—
PuzzleMixBackbone=ResNet50, Training=from scratch2026.065.11——————8.82301
CutMixBackbone=ResNet50, Training=from scratch2026.065.26——————9.080.61
Mixup + CutMixBackbone=DeiT-S, latency [ms]=0.702026.065.31——————7.41—
ResizeMixBackbone=ResNet18, Training=from scratch2026.065.4——————5.040.62
MixupBackbone=DeiT-S, latency [ms]=0.732026.065.42——————6.03—
FMixBackbone=ResNet50, Training=from scratch2026.065.44——————7.656.68
CutMix + decoupledBackbone=ResNet50, Training=from scratch2026.065.48——————7.420.6
OA-CutMix/abs + decoupledBackbone=ResNet50, Training=from scratch2026.065.58——————4.851.01
CutMixBackbone=DeiT-Ti, latency [ms]=0.672026.065.61——————8.75—
VTSTBackbone=ResNet-502025.045.75———59.465.75———
SnapMixBackbone=ResNet50, Training=from scratch2026.065.82——————6.1751.49
Attentive CutMixBackbone=DeiT-Ti, latency [ms]=56.102026.065.87——————7.64—
CutMixBackbone=DeiT-S, latency [ms]=0.672026.065.98——————8.75—
MixupBackbone=ResNet50, Training=from scratch2026.065.99——————12.180.73
TSTBackbone=ResNet-1012025.046.01———61.496.01———
TSTBackbone=ResNet-502025.046.18———59.966.18———
CutMix + decoupledBackbone=ResNet18, Training=from scratch2026.066.47——————7.420.6
SnapMixBackbone=DeiT-S, latency [ms]=39.982026.066.72——————10.15—
OA-CutMix/relBackbone=ResNet50, Training=from scratch2026.066.87——————4.761.02
CutMixBackbone=ResNet18, Training=from scratch2026.066.93——————9.080.61
OA-CutMix/absBackbone=DeiT-S, latency [ms]=1.012026.066.94——————3.27—
OA-CutMix/rel + decoupledBackbone=ResNet50, Training=from scratch2026.067.02——————4.661
FMixBackbone=ResNet18, Training=from scratch2026.067.1——————7.656.68
ResizeMixBackbone=ResNet50, Training=from scratch2026.067.18——————5.040.62
Attentive CutMixBackbone=ResNet18, Training=from scratch2026.067.51——————8.6956.36
OA-CutMix/abs + decoupledBackbone=DeiT-S, latency [ms]=1.012026.067.75——————3.66—
TokenMixBackbone=DeiT-Ti, latency [ms]=39.912026.068.01——————7.52—
Attentive CutMixBackbone=ResNet50, Training=from scratch2026.068.84——————8.6956.36
SnapMixBackbone=ResNet18, Training=from scratch2026.068.9——————6.1751.49
FMixBackbone=DeiT-Ti, latency [ms]=6.802026.068.91——————6.91—
OA-CutMix/relBackbone=DeiT-S, latency [ms]=1.032026.069.15——————3.96—
ResizeMixBackbone=DeiT-S, latency [ms]=1.052026.069.57——————7.44—
PuzzleMixBackbone=ResNet18, Training=from scratch2026.069.7——————8.82301
OA-CutMix/rel + decoupledBackbone=DeiT-S, latency [ms]=1.112026.0610.53——————4.6—
FMixBackbone=DeiT-S, latency [ms]=6.802026.0610.63——————6.91—
ResizeMixBackbone=DeiT-Ti, latency [ms]=1.052026.0611.18——————7.44—
Attentive CutMixBackbone=DeiT-S, latency [ms]=56.102026.0611.74——————7.64—
CutMix + decoupledBackbone=DeiT-S, latency [ms]=0.672026.0613.96——————4.71—
VanillaBackbone=ResNet50, Training=from scratch2026.0615.46——————9.33—
Mixup+CutMixBackbone=ResNet18, Training=from scratch2026.0616.3——————12.250.67
MixupBackbone=ResNet18, Training=from scratch2026.0616.62——————12.180.73
VanillaBackbone=DeiT-Ti2026.0617.01——————14.78—
VanillaBackbone=ResNet18, Training=from scratch2026.0618.26——————9.33—
AlignmixupBackbone=ResNet50, Training=from scratch2026.0620.81——————6.131.98
VanillaBackbone=DeiT-S2026.0630.36——————14.78—
Softmax Temperature TuningTemperature Tuning=true, Outlier Exposure=false2018.12—12962.9—————
Softmax Temperature Tuning + OETemperature Tuning=true, Outlier Exposure=true2018.12—6.94.872.2—————