Image Classification on Digits-DG (leave-one-domain-out)
90.93Average AccuracyALOFT-E
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ALOFT-EArchitecture Category=MLP-like models, Params.=14M2023.03 | 90.93 | 98.45 | 83.35 | 84.55 | 97.37 | |
| ALOFT-SArchitecture Category=MLP-like models, Params.=14M2023.03 | 90.74 | 98.18 | 83.21 | 84.38 | 97.2 | |
| FAMLP-BArchitecture Category=MLP-like models, Params.=25M2023.03 | 90.6 | 98 | 83.3 | 84.1 | 96.9 | |
| Strong BaselineArchitecture Category=MLP-like models, Params.=14M2023.03 | 87.39 | 97.45 | 74.05 | 80.83 | 96.71 | |
| D-GAPProtocol=Leave-one-domain-out evaluation2025.11 | 84.5 | 98.3 | 68.5 | 77.2 | 93.8 | |
| STEAMBackbone=4 conv layers2021.12 | 83.1 | 96.8 | 67.5 | 76 | 92.2 | |
| STEAMArchitecture Category=CNN: ResNet-18, Params.=11M2023.03 | 83.1 | 96.8 | 67.5 | 76 | 92.2 | |
| StableFDGDistribution=single-domain2023.11 | 82.64 | 97.23 | 74.53 | 72.95 | 85.85 | |
| Su et al. (2024)Protocol=Leave-one-domain-out evaluation2025.11 | 82.6 | 97.3 | 65.9 | 74.5 | 92.5 | |
| CIRLEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 82.5 | 96.08 | 69.87 | 76.17 | 87.68 | |
| CIRLArchitecture Category=CNN: ResNet-18, Params.=11M2023.03 | 82.5 | 96.08 | 69.87 | 76.17 | 87.68 | |
| DSUDistribution=single-domain2023.11 | 82.49 | 98.1 | 75.6 | 70.47 | 85.8 | |
| COMENArchitecture Category=CNN: ResNet-18, Params.=11M2023.03 | 82.3 | 97.1 | 67.6 | 75.1 | 91.3 | |
| FedDGDistribution=single-domain2023.11 | 82.25 | 97.97 | 72.13 | 71.03 | 87.87 | |
| SAM (2023)Protocol=Leave-one-domain-out evaluation2025.11 | 82.1 | 97.8 | 66.8 | 73.2 | 90.7 | |
| MixStyleDistribution=single-domain2023.11 | 82.09 | 97.75 | 74.25 | 70.85 | 85.5 | |
| FedBNDistribution=single-domain2023.11 | 81.79 | 97.33 | 72.68 | 71.77 | 85.36 | |
| FedSRDistribution=single-domain2023.11 | 81.55 | 98 | 73 | 68.5 | 86.7 | |
| FACT2021.05 | 81.5 | — | — | — | — | |
| FACTbackbone=vanilla ConvNet2021.05 | 81.5 | 97.9 | 65.6 | 72.4 | 90.3 | |
| FACTEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 81.5 | 97.9 | 65.6 | 72.4 | 90.3 | |
| FACTArchitecture Category=CNN: ResNet-18, Params.=11M2023.03 | 81.5 | 97.9 | 65.6 | 72.4 | 90.3 | |
| FACT (2023)Protocol=Leave-one-domain-out evaluation2025.11 | 81.5 | 97.9 | 65.6 | 72.4 | 90.3 | |
| FedAvgDistribution=single-domain2023.11 | 81.09 | 98.05 | 70.95 | 68.95 | 86.4 | |
| AGST (2025)Protocol=Leave-one-domain-out evaluation2025.11 | 81 | 97.3 | 63.1 | 70.2 | 93.2 | |
| Inference-time Label-Preserving Target ProjectionsBackbone=Architecture from [48]2021.03 | 80.3 | 97.99 | 66.52 | 71.31 | 85.4 | |
| L2A-OTBackbone=four 64-kernel 3x3 convolution layers2020.07 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OT2021.04 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OT2021.05 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OTbackbone=vanilla ConvNet2021.05 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OTBackbone=Architecture from [48]2021.03 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OTBackbone=4 conv layers2021.12 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| L2A-OTEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 78.1 | 96.7 | 63.9 | 68.6 | 83.2 | |
| DDAIGBackbone=4-layer ConvNet, Input Resolution=32x322020.03 | 77.6 | 96.6 | 64.1 | 68.6 | 81 | |
| DDAIG2021.05 | 77.6 | — | — | — | — | |
| DDAIGbackbone=vanilla ConvNet2021.05 | 77.6 | 96.6 | 64.1 | 68.6 | 81 | |
| DDAIGEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 77.6 | 96.6 | 64.1 | 68.6 | 81 | |
| CNN + MixStyleAugmentation=MixStyle2021.04 | 76.5 | 96.5 | 63.5 | 64.7 | 81.2 | |
| MixStyleBackbone=4 conv layers2021.12 | 76.5 | 96.5 | 63.5 | 64.7 | 81.2 | |
| CrossGradBackbone=4-layer ConvNet, Input Resolution=32x322020.03 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGradBackbone=four 64-kernel 3x3 convolution layers2020.07 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGrad2021.04 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGrad2021.05 | 75.8 | — | — | — | — | |
| CrossGradbackbone=vanilla ConvNet2021.05 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGradBackbone=Architecture from [48]2021.03 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGradBackbone=4 conv layers2021.12 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CrossGradEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 75.8 | 96.7 | 61.1 | 65.3 | 80.2 | |
| CNN + DropBlockRegularization=DropBlock2021.04 | 75.3 | 96.2 | 60.5 | 64.1 | 80.2 | |
| MMD-AAEBackbone=4-layer ConvNet, Input Resolution=32x322020.03 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAEBackbone=four 64-kernel 3x3 convolution layers2020.07 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAE2021.04 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAE2021.05 | 74.6 | — | — | — | — | |
| MMD-AAEbackbone=vanilla ConvNet2021.05 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAEBackbone=Architecture from [48]2021.03 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAEBackbone=4 conv layers2021.12 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| MMD-AAEEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 74.6 | 96.5 | 58.4 | 65 | 78.4 | |
| CCSABackbone=4-layer ConvNet, Input Resolution=32x322020.03 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSABackbone=four 64-kernel 3x3 convolution layers2020.07 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSA2021.04 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSA2021.05 | 74.5 | — | — | — | — | |
| CCSAbackbone=vanilla ConvNet2021.05 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSABackbone=Architecture from [48]2021.03 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSABackbone=4 conv layers2021.12 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CCSAEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 74.5 | 95.2 | 58.2 | 65.5 | 79.1 | |
| CNN + CutoutAugmentation=Cutout2021.04 | 74.1 | 95.8 | 58.4 | 61.9 | 80.6 | |
| JiGenBackbone=four 64-kernel 3x3 convolution layers2020.07 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| JiGen2021.04 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| Jigen2021.05 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| Jigenbackbone=vanilla ConvNet2021.05 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| JigenBackbone=Architecture from [48]2021.03 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| JiGenBackbone=4 conv layers2021.12 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| JiGenEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 73.9 | 96.5 | 61.4 | 63.7 | 74 | |
| VanillaBackbone=4-layer ConvNet, Input Resolution=32x322020.03 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| VanillaBackbone=four 64-kernel 3x3 convolution layers2020.07 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| CNN2021.04 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| DeepAll2021.05 | 73.7 | — | — | — | — | |
| DeepAllbackbone=vanilla ConvNet2021.05 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| VanillaBackbone=4 conv layers2021.12 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| DeepAllEvaluation Protocol=leave-one-domain-out, Resolution=32x322022.03 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| DeepAllArchitecture Category=CNN: ResNet-18, Params.=11M2023.03 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| DeepAllProtocol=Leave-one-domain-out evaluation2025.11 | 73.7 | 95.8 | 58.8 | 61.7 | 78.6 | |
| Deep AllBackbone=Architecture from [48]2021.03 | 73.66 | 95.24 | 58.36 | 62.12 | 78.94 | |
| CNN + MixupAugmentation=Mixup2021.04 | 72.7 | 94.2 | 56.5 | 63.3 | 76.7 | |
| CCSTDistribution=single-domain, Augmentation Level=32023.11 | 72.35 | 95.1 | 62.8 | 56.6 | 74.9 | |
| CNN + CutMixAugmentation=CutMix2021.04 | 71.8 | 94.9 | 50.1 | 64.1 | 78.1 | |
| CNN + Manifold MixupAugmentation=Manifold Mixup2021.04 | 71.7 | 92.7 | 53.1 | 64.4 | 76.8 | |
| CNN + Mixup w/o label interpolationAugmentation=Mixup w/o label interpolation2021.04 | 71.2 | 93.7 | 55.2 | 61.6 | 74.4 |