Image Classification on PACS (test)
97.51Average AccuracyWATT-S
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| WATT-SVisual Encoder=ViT-L/142024.06 | 97.51 | 98.68 | 97.9 | 93.8 | 99.64 | — | — | |
| TENTVisual Encoder=ViT-L/142024.06 | 97.41 | 98.83 | 97.74 | 93.51 | 99.54 | — | — | |
| CLIPARTTVisual Encoder=ViT-L/142024.06 | 97.33 | 98.76 | 97.74 | 93.26 | 99.54 | — | — | |
| CLIPVisual Encoder=ViT-L/142024.06 | 97.31 | 98.68 | 97.74 | 93.28 | 99.54 | — | — | |
| DenseModel=Swin-T2024.04 | 97.1 | — | — | — | — | — | — | |
| Smallest weightsModel=Swin-T2024.04 | 96.1 | — | — | — | — | — | — | |
| WATT-SVisual Encoder=ViT-B/162024.06 | 96.08 | 97.66 | 97.51 | 89.56 | 99.58 | — | — | |
| WATT-PVisual Encoder=ViT-B/162024.06 | 96.07 | 97.49 | 97.47 | 89.73 | 99.58 | — | — | |
| Smallest gradientsModel=Swin-T2024.04 | 95.7 | — | — | — | — | — | — | |
| TPTVisual Encoder=ViT-L/142024.06 | 95.66 | 94.82 | 95.65 | 92.72 | 99.44 | — | — | |
| CLIPARTTVisual Encoder=ViT-B/162024.06 | 95.35 | 97.64 | 97.37 | 86.79 | 99.58 | — | — | |
| TENTVisual Encoder=ViT-B/162024.06 | 95.22 | 97.54 | 97.37 | 86.37 | 99.58 | — | — | |
| CLIPVisual Encoder=ViT-B/162024.06 | 95.12 | 97.44 | 97.38 | 86.06 | 99.58 | — | — | |
| NEPENTHEModel=Swin-T2024.04 | 95.1 | — | — | — | — | — | — | |
| DenseModel=ResNet-182024.04 | 94.7 | — | — | — | — | — | — | |
| Group lassoModel=Swin-T2024.04 | 94.3 | — | — | — | — | — | — | |
| IMPModel=Swin-T2024.04 | 93.9 | — | — | — | — | — | — | |
| EGPModel=Swin-T2024.04 | 93.5 | — | — | — | — | — | — | |
| DenseModel=MobileNet-V22024.04 | 93.2 | — | — | — | — | — | — | |
| TPTVisual Encoder=ViT-B/162024.06 | 93.08 | 95.1 | 91.42 | 87.23 | 98.56 | — | — | |
| STEAMBackbone=ResNet-182021.12 | 93 | 94 | 93.7 | 85.1 | 99.3 | — | — | |
| NEPENTHEModel=MobileNet-V22024.04 | 92.2 | — | — | — | — | — | — | |
| Group lassoModel=MobileNet-V22024.04 | 92.1 | — | — | — | — | — | — | |
| ALOFT-EBackbone=GFNet-H-Ti, Params.=14M2023.03 | 91.58 | 92.24 | 87.84 | 87.38 | 98.86 | — | — | |
| IMPModel=MobileNet-V22024.04 | 91.4 | — | — | — | — | — | — | |
| ALOFT-SBackbone=GFNet-H-Ti, Params.=14M2023.03 | 90.88 | 91.7 | 85.49 | 87.58 | 98.76 | — | — | |
| CIRLBackbone=ResNet-50, Params.=23M2023.03 | 90.12 | 90.67 | 84.3 | 87.68 | 97.84 | — | — | |
| FAMLP-SBackbone=MLP-like, Params.=44M2023.03 | 90.12 | 92.63 | 87.03 | 82.69 | 98.14 | — | — | |
| NEPENTHEModel=ResNet-182024.04 | 90.1 | — | — | — | — | — | — | |
| IMPModel=ResNet-182024.04 | 89.8 | — | — | — | — | — | — | |
| CMSSBackbone=ResNet-182021.12 | 89.5 | 88.6 | 90.4 | 82 | 96.9 | — | — | |
| MVDGBackbone=ResNet-50, Params.=23M2023.03 | 89.33 | 89.31 | 84.22 | 86.36 | 97.43 | — | — | |
| FAMLP-BBackbone=MLP-like, Params.=25M2023.03 | 89.19 | 92.06 | 82.49 | 84.09 | 98.1 | — | — | |
| StyleNeophileBackbone=ResNet-50, Params.=23M2023.03 | 89.11 | 90.35 | 84.2 | 85.18 | 96.73 | — | — | |
| M3SDABackbone=ResNet-182021.12 | 88.3 | 89.3 | 89.9 | 76.7 | 97.3 | — | — | |
| ViP-SBackbone=MLP-like, Params.=25M2023.03 | 88.27 | 88.09 | 84.22 | 82.41 | 98.38 | — | — | |
| I2-ADRBackbone=ResNet-50, Params.=23M2023.03 | 88.2 | 88.5 | 83.2 | 85.8 | 95.2 | — | — | |
| FACTBackbone=ResNet-50, Params.=23M2023.03 | 88.15 | 89.63 | 81.77 | 84.46 | 96.75 | — | — | |
| Core-tuningPre-training=MoCo-v2, Fine-tuning=Core-tuning, Backbone=ResNet-502021.02 | 88.08 | 87.31 | 84.06 | 83.43 | 97.53 | — | — | |
| Strong Baseline (GFNet)Backbone=GFNet-H-Ti, Params.=14M2023.03 | 87.76 | 89.37 | 84.74 | 79.01 | 97.94 | — | — | |
| Single sample generalizationBackbone=ResNet-502022.02 | 87.51 | 88.09 | 83.83 | 80.21 | 97.88 | — | — | |
| MCDBackbone=ResNet-182021.12 | 87 | 88.7 | 88.9 | 73.9 | 96.4 | — | — | |
| EFDMixBackbone=ResNet-50, Params.=23M2023.03 | 86.9 | 90.6 | 82.5 | 76.4 | 98.1 | — | — | |
| Seo et al. (2020)Backbone=ResNet-502022.02 | 86.64 | 87.04 | 80.62 | 82.9 | 95.99 | — | — | |
| MVDGBackbone=ResNet-18, Params.=11M2023.03 | 86.56 | 85.62 | 79.98 | 85.08 | 95.54 | — | — | |
| XDEDBackbone=ResNet-18, Params.=11M2023.03 | 86.4 | 85.6 | 84.2 | 79.1 | 96.5 | — | — | |
| CIRLBackbone=ResNet-18, Params.=11M2023.03 | 86.32 | 86.08 | 80.59 | 82.67 | 95.93 | — | — | |
| Wang et al. (2021)Backbone=ResNet-50, reimplemented=true2022.02 | 86.23 | 86.3 | 82.53 | 78.11 | 97.96 | — | — | |
| COMENBackbone=ResNet-18, Params.=11M2023.03 | 85.7 | 82.6 | 81 | 84.5 | 94.6 | — | — | |
| MoCo-v2 (CE-Con)Pre-training=MoCo-v2, Fine-tuning=CE-Con, Backbone=ResNet-502021.02 | 85.65 | 85.11 | 81.77 | 80.12 | 95.58 | — | — | |
| MBDGModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 85.6 | 80.6 | 79.3 | 85.2 | 97 | — | — | |
| I2-ADRBackbone=ResNet-18, Params.=11M2023.03 | 85.6 | 82.9 | 80.8 | 83.5 | 95 | — | — | |
| Gulrajani & Lopez-Paz (2020)Backbone=ResNet-502022.02 | 85.5 | 84.7 | 80.8 | 79.3 | 97.2 | — | — | |
| StyleNeophileBackbone=ResNet-18, Params.=11M2023.03 | 85.47 | 84.41 | 79.25 | 83.27 | 94.93 | — | — | |
| Zhao et al. (2020)Backbone=ResNet-502022.02 | 85.34 | 87.51 | 79.31 | 76.3 | 98.25 | — | — | |
| Supervised (CE)Pre-training=Supervised, Fine-tuning=Cross-Entropy (CE), Backbone=ResNet-502021.02 | 85.11 | 83.65 | 79.21 | 81.46 | 96.11 | — | — | |
| Seo et al. (2020)Backbone=ResNet-182022.02 | 85.11 | 84.67 | 77.65 | 82.23 | 95.87 | — | — | |
| Zhou et al. (2020a)Backbone=ResNet-50, reimplemented=true2022.02 | 84.9 | 85.21 | 80.33 | 76.53 | 97.55 | — | — | |
| FACTBackbone=ResNet-18, Params.=11M2023.03 | 84.51 | 85.37 | 78.38 | 79.15 | 95.15 | — | — | |
| Dubey et al. (2021)Backbone=ResNet-502022.02 | 84.5 | — | — | — | — | — | — | |
| EGPModel=ResNet-182024.04 | 84.3 | — | — | — | — | — | — | |
| L2DBackbone=ResNet-18, Domain Identifier=false2021.08 | 84.27 | 81.44 | 79.56 | 80.58 | 95.51 | — | — | |
| gMLP-SBackbone=MLP-like, Params.=20M2023.03 | 84.23 | 86.72 | 80.8 | 72.13 | 97.54 | — | — | |
| NAS-OODBackbone=Searched, Params (M)=3.36 (Average)2021.09 | 84.23 | 83.74 | 79.69 | 77.27 | 96.23 | — | — | |
| Single sample generalizationBackbone=ResNet-182022.02 | 84.15 | 82.02 | 79.73 | 78.96 | 95.87 | — | — | |
| RepMLPBackbone=MLP-like, Params.=38M2023.03 | 84.12 | 82.28 | 78.8 | 79.49 | 95.93 | — | — | |
| EFDMixBackbone=ResNet-18, Params.=11M2023.03 | 83.9 | 83.9 | 79.4 | 75 | 96.8 | — | — | |
| WBNBackbone=ResNet-182021.12 | 83.8 | 89.9 | 89.7 | 58 | 97.4 | — | — | |
| MTLModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 83.7 | 85.6 | 78.9 | 73.1 | 97.1 | — | — | |
| Zhou et al. (2020b)Backbone=ResNet-182022.02 | 83.7 | 84.1 | 78.8 | 75.9 | 96.1 | — | — | |
| SelfRegBackbone=ResNet182021.04 | 83.62 | 82.34 | 78.43 | 77.47 | 96.22 | — | — | |
| GroupDROModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 83.5 | 84.4 | 77.3 | 75.6 | 96.8 | — | — | |
| ResMLP-SBackbone=MLP-like, Params.=40M2023.03 | 83.46 | 85.5 | 78.63 | 72.64 | 97.07 | — | — | |
| MixupModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 83.2 | 85.2 | 77 | 73.9 | 96.8 | — | — | |
| MMDModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 83.2 | 84.9 | 75.1 | 76.5 | 96.1 | — | — | |
| DDAIGBackbone=ResNet-182020.07 | 83.1 | 84.2 | 78.1 | 74.7 | 95.3 | — | — | |
| Wang et al. (2021)Backbone=ResNet-18, reimplemented=true2022.02 | 83.09 | 81.55 | 77.67 | 77.64 | 95.49 | — | — | |
| ERMModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 83 | 83.2 | 76.8 | 74.8 | 97.2 | — | — | |
| MLDGModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 82.9 | 81.4 | 77.9 | 76.1 | 96.2 | — | — | |
| RotationBackbone=ResNet-18, protocol=max target accuracy2020.07 | 82.85 | 82.4 | 75.27 | 77.2 | 96.53 | — | — | |
| Zhou et al. (2020a)Backbone=ResNet-182022.02 | 82.83 | 83.3 | 78.2 | 73.6 | 96.2 | — | — | |
| L2A-OTBackbone=ResNet-18, Params (M)=11.72021.09 | 82.8 | 83.3 | 78.2 | 73.6 | 96.2 | — | — | |
| Dou et al. (2019)Backbone=ResNet-502022.02 | 82.67 | 82.89 | 80.49 | 72.29 | 95.01 | — | — | |
| CORALModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 82.6 | 80.5 | 74.5 | 78.6 | 96.8 | — | — | |
| RSCModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 82.6 | 83.7 | 82.9 | 68.1 | 95.6 | — | — | |
| DecAugBackbone=ResNet-18, Params (M)=11.72021.09 | 82.39 | 79 | 79.61 | 75.64 | 95.33 | — | — | |
| Jigsaw+RotationBackbone=ResNet-18, protocol=max target accuracy2020.07 | 82.33 | 81.4 | 75.03 | 76.47 | 96.4 | — | — | |
| SagNetModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 82.3 | 81.1 | 75.4 | 77.2 | 95.7 | — | — | |
| FDSE2025.03 | 82.17 | — | — | — | — | 83.81 | — | |
| RSCBackbone=ResNet182021.04 | 82.1 | 79.88 | 76.87 | 77.11 | 94.56 | — | — | |
| MEADABackbone=ResNet-18, Domain Identifier=false2021.08 | 82.1 | 78.61 | 78.65 | 75.59 | 95.57 | — | — | |
| Jigsaw+RotationBackbone=ResNet-18, alpha_J=0.7, alpha_R=0.7, beta=0.82020.07 | 81.87 | 81.07 | 74.13 | 76.17 | 96.1 | — | — | |
| MMLDBackbone=ResNet-182020.07 | 81.83 | 81.28 | 77.16 | 72.29 | 96.09 | — | — | |
| MMLDBackbone=ResNet-18, Domain Identifier=false2021.08 | 81.83 | 81.28 | 77.16 | 72.29 | 96.09 | — | — | |
| MetaRegBackbone=ResNet-182020.07 | 81.7 | 83.7 | 77.2 | 70.3 | 95.5 | — | — | |
| ARMModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 81.7 | 85.9 | 73.3 | 72.1 | 95.6 | — | — | |
| CuMixBackbone=ResNet-18, Params (M)=11.72021.09 | 81.6 | 82.3 | 76.5 | 72.6 | 95.1 | — | — | |
| Epi-FCRBackbone=ResNet-182020.07 | 81.5 | 82.1 | 77 | 73 | 93.9 | — | — | |
| IRMModel selection=hold-one-out cross-validation, Input resolution=224x2242021.02 | 81.5 | 81.7 | 77 | 71.1 | 96.3 | — | — | |
| DANNBackbone=ResNet-182021.12 | 81.5 | 81.9 | 77.5 | 74.6 | 91.8 | — | — |