Domain Generalization on DomainBed
79.6Average AccuracyUniDG + CORAL + ConvNeXt-B
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| UniDG + CORAL + ConvNeXt-BBackbone=ConvNeXt-B, Pre-training=ImageNet2023.10 | 79.6 | 95.6 | 84.5 | — | 69.6 | 59.5 | 88.9 | — | — | |
| UniDGBackbone=ViT-B/16, Pre-training=CLIP2023.10 | 78.6 | 96.7 | 86.3 | — | 62.4 | 61.3 | 86.2 | — | — | |
| CAR-FTBackbone=ViT-B/16, Pre-training=CLIP2023.10 | 78.5 | 96.8 | 85.5 | — | 61.9 | 62.5 | 85.7 | — | — | |
| GMDGEnsemble=true, Oracle model=RegNetY-16GF, SWAD=true2024.02 | 78.2 | 97.9 | 82.2 | — | 65 | 61.3 | 84.7 | — | — | |
| SIMPLE++Ensemble=true, Oracle model=Multiple2024.02 | 78.1 | 99 | 82.7 | — | 59 | 61.9 | 87.7 | — | — | |
| MIROEnsemble=true, Oracle model=RegNetY-16GF, SWAD=true2024.02 | 77.3 | 96.8 | 81.7 | — | 64.3 | 60.7 | 83.3 | — | — | |
| OracleBackbone=ViT-b162023.06 | 76.16 | — | — | — | — | — | — | — | — | |
| GMDGEnsemble=false, Oracle model=RegNetY-16GF2024.02 | 75.1 | 97.3 | 82.4 | — | 60.7 | 54.6 | 80.8 | — | — | |
| DPLBackbone=ViT-B/16, Pre-training=CLIP2023.10 | 75 | 97.3 | 84.3 | — | 52.6 | 56.7 | 84.2 | — | — | |
| SEDGE+Model Pool=Model Pool-B, Algorithm Category=Ensemble2022.03 | 74.1 | 96.1 | 82.2 | — | 56.8 | 54.7 | 80.7 | — | — | |
| MIROBackbone=RegNetY-16GF, Pre-training=SWAG2023.10 | 74.1 | 97.4 | 79.9 | — | 58.9 | 53.8 | 80.4 | — | — | |
| NAC-MEBackbone=ViT-b162023.06 | 74.1 | — | — | — | — | — | — | — | 41.46 | |
| MIROEnsemble=false, Oracle model=RegNetY-16GF2024.02 | 74.1 | 97.4 | 79.9 | — | 58.9 | 53.8 | 80.4 | — | — | |
| MIROBackbone=ViT-B/16, Pre-training=CLIP2023.10 | 73.7 | 95.6 | 82.2 | — | 54.3 | 54 | 82.5 | — | — | |
| ValidationBackbone=ViT-b162023.06 | 73.03 | — | — | — | — | — | — | — | 29.85 | |
| EoA+Model Pool=Model Pool-B, Algorithm Category=Ensemble2022.03 | 72.7 | 93.2 | 80.4 | — | 55.2 | 54.6 | 80.2 | — | — | |
| ERMBackbone=ViT-B/16, Pre-training=CLIP2023.10 | 72.2 | 93.7 | 82.7 | — | 52.3 | 53.8 | 78.5 | — | — | |
| SIMPLEEnsemble=true, Oracle model=Multiple2024.02 | 72 | 88.6 | 79.9 | — | 57.6 | 49.2 | 84.6 | — | — | |
| SAGM + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 71.4 | 86.6 | 80 | — | 48.8 | 45 | 70.1 | — | — | |
| UniDG + MIROBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 70.8 | 90.4 | 84.1 | — | 54.4 | 52.6 | 72.5 | — | — | |
| OracleBackbone=ResNet-502023.06 | 70.65 | — | — | — | — | — | — | — | — | |
| CORAL + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 70.3 | 86.2 | 78.8 | — | 47.6 | 41.5 | 68.7 | — | — | |
| GSAM + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 70.3 | 85.9 | 79.1 | — | 47 | 44.6 | 69.3 | — | — | |
| GGA-LAlgorithm=GGA-L2025.02 | 70.3 | 86.5 | 78.4 | — | 49.8 | 44.5 | 66.5 | 65.1 | — | |
| SagNet + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 70.2 | 86.3 | 77.8 | — | 48.6 | 40.3 | 68.1 | — | — | |
| GGAAlgorithm=GGA2025.02 | 70.2 | 86.4 | 78.7 | — | 48.5 | 44.5 | 67 | 65 | — | |
| Mixup + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 69.5 | 84.6 | 77.4 | — | 47.9 | 39.2 | 68.1 | — | — | |
| SAM + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 69.5 | 85.8 | 79.4 | — | 43.3 | 44.3 | 69.6 | — | — | |
| SEDGEModel Pool=Model Pool-A, Algorithm Category=Ensemble2022.03 | 69.4 | 84.1 | 79.8 | — | 56.8 | 46.3 | 79.9 | — | — | |
| Fish + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 69.3 | 85.5 | 77.8 | — | 45.1 | 42.7 | 68.6 | — | — | |
| UniDG + CORALBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 69.3 | 89.2 | 82.1 | — | 53 | 51.3 | 70.6 | — | — | |
| MLDG + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 69.2 | 84.9 | 77.2 | — | 47.7 | 41.2 | 66.8 | — | — | |
| VREX + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 69 | 84.9 | 78.3 | — | 46.4 | 33.6 | 66.4 | — | — | |
| ERMAlgorithm=ERM2025.02 | 68.9 | 85.5 | 77.3 | — | 46.1 | 43.8 | 66.5 | 63.9 | — | |
| RSC + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 68.6 | 85.2 | 77.1 | — | 46.6 | 38.9 | 65.5 | — | — | |
| NAC-MEBackbone=ResNet-502023.06 | 68.51 | — | — | — | — | — | — | — | 50.02 | |
| IRM + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 68.5 | 83.5 | 78.5 | — | 47.6 | 33.9 | 64.3 | — | — | |
| MTL + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 68.5 | 84.6 | 77.2 | — | 45.6 | 40.6 | 66.4 | — | — | |
| UniDGBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 68.5 | 89 | 81.6 | — | 52.9 | 50.2 | 68.9 | — | — | |
| EoQModels=5, Size=1.1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 68.4 | 89.3 | 79.5 | 72.3 | 53.2 | 47.9 | — | — | — | |
| ValidationBackbone=ResNet-502023.06 | 68.34 | — | — | — | — | — | — | — | 48.74 | |
| ARM + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 68.3 | 85.1 | 77.6 | — | 45.5 | 35.5 | 64.8 | — | — | |
| AND-mask + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 68.2 | 84.4 | 78.1 | — | 44.6 | 37.2 | 65.6 | — | — | |
| GMDGEnsemble=true, Oracle model=ResNet-50, SWAD=true2024.02 | 68.2 | 88.4 | 79.6 | — | 53 | 47.3 | 72.5 | — | — | |
| MIROEnsemble=true, Oracle model=ResNet-50, SWAD=true2024.02 | 68.1 | 88.4 | 79.6 | — | 52.9 | 47 | 72.4 | — | — | |
| DIWAModels=60, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 68 | 89 | 78.6 | 72.8 | 51.9 | 47.7 | — | — | — | |
| EoAModels=6, Size=6x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 68 | 88.6 | 79.1 | 72.5 | 52.3 | 47.4 | — | — | — | |
| EoAModel Pool=Model Pool-A, Algorithm Category=Ensemble2022.03 | 68 | 88.6 | 79.1 | — | 52.3 | 47.4 | 72.5 | — | — | |
| ERMBackbone=RegNetY-16GF, Pre-training=SWAG2023.10 | 68 | 89.6 | 78.6 | — | 51.4 | 48.5 | 71.9 | — | — | |
| SWAD + DARTBackbone=ResNet-50, Number of reruns=32023.02 | 67.9 | 88.9 | 80.3 | — | 51.3 | 47.1 | 71.9 | — | — | |
| GMOE-S/16Backbone=ViT-S/16, Pre-training=ImageNet2023.10 | 67.9 | 88.1 | 80.2 | — | 48.5 | 48.7 | 74.2 | — | — | |
| MMD + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 67.7 | 84.7 | 77.5 | — | 42.2 | 23.4 | 66.3 | — | — | |
| GroupDRO + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 67.6 | 84.1 | 76.7 | — | 43.2 | 33.3 | 66 | — | — | |
| SWADModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 66.9 | 88.1 | 79.1 | 70.6 | 50 | 46.5 | — | — | — | |
| Mixstyle + GGAOptimization=GGA, Base Architecture=ResNet-502025.02 | 66.9 | 85.2 | 77.9 | — | 44 | 34 | 60.4 | — | — | |
| SWADModel Pool=Model Pool-A, Algorithm Category=Ensemble2022.03 | 66.9 | 88.1 | 79.1 | — | 50 | 46.5 | 70.6 | — | — | |
| SWADBackbone=ResNet-50, Number of reruns=32023.02 | 66.9 | 88.1 | 79.1 | — | 50 | 46.5 | 70.6 | — | — | |
| SWADBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 66.9 | 88.1 | 79.1 | — | 50 | 46.5 | 70.6 | — | — | |
| ERM Ens.Models=6, Size=6x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 66.8 | 87.6 | 78.5 | 70.8 | 49.2 | 47.7 | — | — | — | |
| AdaNPCBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 66.5 | 88.9 | 80.2 | — | 54 | 43.1 | 66.3 | — | — | |
| GMDGEnsemble=false, Oracle model=ResNet-502024.02 | 66.3 | 85.6 | 79.2 | — | 51.1 | 44.6 | 70.7 | — | — | |
| QT-DOGModels=1, Size=0.22x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 66.2 | 87.8 | 78.4 | 68.9 | 50.8 | 45.1 | — | — | — | |
| DARTModels=4-6, Size=4x-6x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 66.1 | 78.5 | 87.3 | 70.1 | 48.7 | 45.8 | — | — | — | |
| ERM + DARTBackbone=ResNet-50, Number of reruns=32023.02 | 66.1 | 87.3 | 78.5 | — | 48.7 | 45.8 | 70.1 | — | — | |
| MIROModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 65.9 | 85.4 | 79 | 70.5 | 50.4 | 44.3 | — | — | — | |
| MIROBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 65.9 | 85.4 | 79 | — | 50.4 | 44.3 | 70.5 | — | — | |
| MIROEnsemble=false, Oracle model=ResNet-502024.02 | 65.9 | 85.4 | 79 | — | 50.4 | 44.3 | 70.5 | — | — | |
| FishrModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 65.7 | 85.5 | 77.8 | 67.8 | 47.4 | 41.7 | — | — | — | |
| ERMBackbone=ViT-S/16, Pre-training=ImageNet2023.10 | 65.5 | 86.2 | 79.7 | — | 42 | 47.3 | 72.2 | — | — | |
| mDSDIEnsemble=false2024.02 | 65.1 | 86.2 | 79 | — | 48.1 | 42.8 | 69.2 | — | — | |
| CCFPModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 64.8 | 86.6 | 78.9 | 68.9 | 48.6 | 41.2 | — | — | — | |
| OracleBackbone=ResNet-182023.06 | 64.72 | — | — | — | — | — | — | — | — | |
| CORALEnsemble=false2024.02 | 64.5 | 86.2 | 78.8 | — | 47.6 | 41.5 | 68.7 | — | — | |
| SagNetModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 64.2 | 86.3 | 77.8 | 68.1 | 48.6 | 40.3 | — | — | — | |
| FishrBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 64.2 | 85.5 | 77.8 | — | 47.4 | 41.7 | 68.6 | — | — | |
| ERMEnsemble=false2024.02 | 64.2 | 84.2 | 77.3 | — | 47.8 | 44 | 67.6 | — | — | |
| SagNetEnsemble=false2024.02 | 64.2 | 86.3 | 77.8 | — | 48.6 | 40.3 | 68.1 | — | — | |
| SelfRegEnsemble=false2024.02 | 64.2 | 85.6 | 77.8 | — | 47 | 42.8 | 67.9 | — | — | |
| CORALModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 64.1 | 86 | 77.7 | 68.6 | 46.4 | 41.8 | — | — | — | |
| CORALModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 64.1 | 86 | 77.7 | — | 46.4 | 41.8 | 68.6 | — | — | |
| CORALBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 64.1 | 86 | 77.7 | — | 46.4 | 41.8 | 68.6 | — | — | |
| DANNBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 64 | 84.6 | 78.7 | — | 46.4 | 41.8 | 68.6 | — | — | |
| MixupModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 63.9 | 84.3 | 77.7 | 69 | 48.9 | 39.6 | — | — | — | |
| FishModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 63.9 | 85.5 | 77.8 | 68.6 | 45.1 | 42.7 | — | — | — | |
| FishModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 63.9 | 85.5 | 77.8 | — | 45.1 | 42.7 | 68.6 | — | — | |
| FISHBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 63.9 | — | — | — | — | — | — | — | — | |
| SagNetBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 63.9 | 85.5 | 77.8 | — | 45.1 | 42.7 | 68.6 | — | — | |
| FishEnsemble=false2024.02 | 63.9 | 85.5 | 77.8 | — | 45.1 | 42.7 | 68.6 | — | — | |
| ERMModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 63.8 | 84.7 | 77.4 | 67.5 | 46.2 | 41.2 | — | — | — | |
| ERMModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 63.8 | 85.7 | 77.4 | — | 47.2 | 41.2 | 67.5 | — | — | |
| ERMBackbone=ResNet-50, Pre-training=ImageNet2023.10 | 63.8 | 85.7 | 77.4 | — | 47.2 | 41.2 | 67.5 | — | — | |
| MLDGModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 63.6 | 84.8 | 77.1 | 68.2 | 46.1 | 41.8 | — | — | — | |
| MMDModels=1, Size=1x, Backbone=ResNet-50, Pre-trained=ImageNet2024.10 | 63.6 | 85 | 76.7 | 67.7 | 49.3 | 39.4 | — | — | — | |
| MLDGModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 63.6 | 84.8 | 77.1 | — | 46.1 | 41.8 | 68.2 | — | — | |
| MMDModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 63.6 | 85 | 76.7 | — | 49.3 | 39.4 | 67.7 | — | — | |
| MLDGEnsemble=false2024.02 | 63.6 | 84.9 | 77.2 | — | 47.7 | 41.2 | 66.8 | — | — | |
| OracleBackbone=ViT-t162023.06 | 63.46 | — | — | — | — | — | — | — | — | |
| ERMBackbone=ResNet-50, Number of reruns=32023.02 | 63.3 | 85.5 | 77.5 | — | 46.1 | 40.9 | 66.5 | — | — | |
| DANNModel Pool=Model Pool-A, Algorithm Category=Non-ensemble2022.03 | 63.1 | 84.6 | 78.7 | — | 48.4 | 38.4 | 65.4 | — | — | |
| MTLEnsemble=false2024.02 | 62.9 | 84.6 | 77.2 | — | 45.6 | 40.6 | 66.4 | — | — |