Image Classification on PathMNIST v1 (test)
92.02AccuracyReproduced
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ReproducedBackbone=ConvNet, Resolution=256x256, Images per class=~10k, Data Regime=Real2024.06 | 92.02 | 88.76 | 99.25 | |
| Reproduced [30,31]Backbone=ConvNet, Resolution=64x64, Images per class=~10k, Data Regime=Real2024.06 | 91.44 | 88.36 | 99.23 | |
| Reproduced [30,31]Backbone=ResNet18, Resolution=64x64, Images per class=~10k, Data Regime=Real2024.06 | 91.25 | 87.95 | 99 | |
| ReproducedBackbone=ResNet18, Resolution=256x256, Images per class=~10k, Data Regime=Real2024.06 | 90.96 | 87.27 | 98.92 | |
| Rate-InFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.052024.12 | 90 | — | 98 | |
| ReproducedBackbone=ResNet18, Resolution=224x224, Images per class=~10k, Data Regime=Real2024.06 | 88.45 | — | 98.35 | |
| Rate-InFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.12024.12 | 88 | — | 97 | |
| ScheduledFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.052024.12 | 88 | — | 97 | |
| ConstantFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.052024.12 | 87 | — | 96 | |
| ActivationFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.052024.12 | 87 | — | 96 | |
| ReproducedBackbone=ResNet18, Resolution=64x64, Images per class=1000, Data Regime=Synthetic2024.06 | 86.18 | 80.8 | 98.15 | |
| SOTA reference [30,31,7,20]Backbone=ResNet18, Resolution=64x64, Images per class=~10k, Data Regime=Real2024.06 | 85.28 | — | 97.33 | |
| InfoDistBackbone=ResNet18, Resolution=256x256, Images per class=100, Data Regime=real distilled2024.06 | 83.9 | 77.47 | 97.77 | |
| InfoDistBackbone=ConvNet, Resolution=256x256, Images per class=100, Data Regime=real distilled2024.06 | 81.17 | 73.81 | 96.73 | |
| ReproducedBackbone=ConvNet, Resolution=64x64, Images per class=1000, Data Regime=Synthetic2024.06 | 79.87 | 73.59 | 96.74 | |
| InfoDistBackbone=ResNet18, Resolution=64x64, Images per class=100, Data Regime=real distilled2024.06 | 78.5 | 71.77 | 96.47 | |
| ScheduledFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.12024.12 | 78 | — | 93 | |
| InfoDistBackbone=ResNet18, Resolution=256x256, Images per class=100, Data Regime=syn distilled2024.06 | 77.79 | 72.53 | 94.59 | |
| InfoDistBackbone=ResNet18, Resolution=64x64, Images per class=100, Data Regime=syn distilled2024.06 | 77.48 | 70.8 | 96.84 | |
| InfoDistBackbone=ConvNet, Resolution=64x64, Images per class=100, Data Regime=real distilled2024.06 | 77.34 | 69.45 | 94.98 | |
| ActivationFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.12024.12 | 76 | — | 92 | |
| Rate-InFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.22024.12 | 75 | — | 91 | |
| ConstantFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.12024.12 | 75 | — | 91 | |
| RandomBackbone=ResNet18, Resolution=64x64, Images per class=100, Data Regime=syn distilled2024.06 | 72.82 | 64.74 | 94.33 | |
| RandomBackbone=ResNet18, Resolution=256x256, Images per class=100, Data Regime=syn distilled2024.06 | 71.82 | 63.88 | 94.99 | |
| InfoDistBackbone=ConvNet, Resolution=64x64, Images per class=100, Data Regime=syn distilled2024.06 | 69.79 | 63.01 | 91.15 | |
| InfoDistBackbone=ConvNet, Resolution=256x256, Images per class=100, Data Regime=syn distilled2024.06 | 65.45 | 57.47 | 91.45 | |
| RandomBackbone=ConvNet, Resolution=64x64, Images per class=100, Data Regime=syn distilled2024.06 | 61.52 | 54.29 | 89.31 | |
| RandomBackbone=ConvNet, Resolution=256x256, Images per class=100, Data Regime=syn distilled2024.06 | 58.29 | 50.95 | 89.97 | |
| ScheduledFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.22024.12 | 47 | — | 65 | |
| GLaDBackbone=ResNet18, Resolution=64x64, Images per class=1, Data Regime=syn distilled2024.06 | 46.68 | 34.63 | 75.97 | |
| GLaDBackbone=ResNet18, Resolution=256x256, Images per class=1, Data Regime=syn distilled2024.06 | 45.7 | 35.59 | 82.24 | |
| ActivationFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.22024.12 | 44 | — | 60 | |
| ConstantFull Model Accuracy=0.9, Dropout Rate (ρ,ε)=0.22024.12 | 41 | — | 56 | |
| GLaDBackbone=ConvNet, Resolution=64x64, Images per class=1, Data Regime=syn distilled2024.06 | 40.49 | 28.46 | 74.45 | |
| GLaDBackbone=ConvNet, Resolution=256x256, Images per class=1, Data Regime=syn distilled2024.06 | 38.81 | 29.44 | 70.26 |