Domain Generalization on Office-Home (Target Domain Performance)
82.6Average AccuracyDCT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DCTBackbone=RegnetY 16GF2023.03 | 82.6 | 81.3 | 69.9 | 89.1 | 89.9 | |
| MIROBackbone=RegnetY 16GF2023.03 | 80.4 | — | — | — | — | |
| ERM+SWADBackbone=RegnetY 16GF2023.03 | 80 | — | — | — | — | |
| START-MParams.=22M, Architecture=SSM-based models2024.10 | 77.09 | 75.15 | 62.04 | 85.31 | 85.84 | |
| START-XParams.=22M, Architecture=SSM-based models2024.10 | 77.07 | 75.48 | 62.06 | 85.24 | 85.47 | |
| DGMambaParams.=22M, Architecture=SSM-based models2024.10 | 77 | 76.2 | 61.8 | 83.9 | 86.1 | |
| Strong BaselineParams.=22M, Architecture=SSM-based models2024.10 | 76.43 | 75.06 | 60.48 | 84.71 | 85.45 | |
| DiPrompTParadigm=FL + Adapter/Prompt, Backbone=ResNet-502024.03 | 76.18 | 74.21 | 58.9 | 85.51 | 86.12 | |
| PromptFLParadigm=FL + Adapter/Prompt, Backbone=ResNet-502024.03 | 73.99 | 72.18 | 56.88 | 82.13 | 84.78 | |
| FedCLIPParadigm=FL + Adapter/Prompt, Backbone=ResNet-502024.03 | 73.76 | 71.03 | 56.13 | 83.76 | 84.14 | |
| ViP-SParams.=25M, Architecture=ViT-based or MLP-like models2024.10 | 73.38 | 69.55 | 61.51 | 79.34 | 83.11 | |
| EoAParams.=23M, Architecture=CNN: ResNet-502024.10 | 72.5 | 69.1 | 59.8 | 79.5 | 81.5 | |
| GMoE-SParams.=34M, Architecture=ViT-based or MLP-like models2024.10 | 72.4 | 69.3 | 58 | 79.8 | 82.6 | |
| Ours w/ MixStyleMixStyle=true2022.05 | 72.31 | 69.43 | 59.4 | 78.56 | 81.85 | |
| OursMixStyle=false2022.05 | 72.13 | 69.39 | 57.15 | 80.2 | 81.79 | |
| ERMBackbone=RegnetY 16GF2023.03 | 71.9 | — | — | — | — | |
| iDAGParams.=23M, Architecture=CNN: ResNet-502024.10 | 71.8 | 68.2 | 57.9 | 79.7 | 81.4 | |
| DomainDropParams.=23M, Architecture=CNN: ResNet-502024.10 | 71.75 | 67.33 | 60.39 | 79.05 | 80.22 | |
| PCLParadigm=Centralized Learning, Backbone=ResNet-502024.03 | 71.6 | 67.3 | 59.9 | 78.7 | 80.7 | |
| PCLParams.=23M, Architecture=CNN: ResNet-502024.10 | 71.6 | 67.3 | 59.9 | 78.7 | 80.7 | |
| AGFAParams.=23M, Architecture=CNN: ResNet-502024.10 | 71.5 | 67.5 | 58.5 | 79.3 | 80.7 | |
| SDViTParams.=22M, Architecture=ViT-based or MLP-like models2024.10 | 71.5 | 68.3 | 56.3 | 79.5 | 81.8 | |
| I2ADRParadigm=Centralized Learning, Backbone=ResNet-502024.03 | 71.4 | 70.3 | 55.1 | 80.7 | 79.2 | |
| GMDGParams.=23M, Architecture=CNN: ResNet-502024.10 | 70.7 | 68.9 | 56.2 | 79.9 | 82 | |
| SWAD2022.05 | 70.6 | 66.1 | 57.7 | 78.4 | 80.2 | |
| SWADParadigm=Centralized Learning, Backbone=ResNet-502024.03 | 70.6 | 66.1 | 59.9 | 78.7 | 80.7 | |
| MIROBackbone=ResNet-502023.03 | 70.5 | 67.5 | 54.6 | 78 | 81.6 | |
| DCTBackbone=ResNet-502023.03 | 70.3 | 67 | 54.9 | 78 | 81.1 | |
| SAGMParams.=23M, Architecture=CNN: ResNet-502024.10 | 70.1 | 65.4 | 57 | 78 | 80 | |
| MADGParams.=23M, Architecture=CNN: ResNet-502024.10 | 70.1 | 67.6 | 54.1 | 78.4 | 80.3 | |
| MLP-BParams.=59M, Architecture=ViT-based or MLP-like models2024.10 | 69.33 | 63.45 | 56.31 | 77.81 | 79.76 | |
| PGradParams.=23M, Architecture=CNN: ResNet-502024.10 | 69.3 | 64.7 | 56 | 77.4 | 78.9 | |
| CCFPParams.=23M, Architecture=CNN: ResNet-502024.10 | 68.9 | 63.7 | 55.5 | 77.2 | 79.2 | |
| CORALBackbone=ResNet-502023.03 | 68.7 | 65.3 | 54.4 | 76.5 | 78.4 | |
| FedGMAParadigm=FL+DG, Backbone=ResNet-502024.03 | 68.25 | 65.34 | 52.11 | 75.98 | 79.59 | |
| SagNetBackbone=ResNet-502023.03 | 68.1 | 63.4 | 54.8 | 75.8 | 78.3 | |
| PCL (w/o SWAD)Backbone=ResNet-502023.03 | 67.9 | 64.1 | 53.9 | 75.3 | 78.4 | |
| ERMBackbone=ResNet-502023.03 | 67.6 | 63.1 | 51.9 | 77.2 | 78.1 | |
| ERM2022.05 | 67.5 | 62.7 | 53.4 | 76.5 | 77.3 | |
| EQRMParams.=23M, Architecture=CNN: ResNet-502024.10 | 67.5 | 60.5 | 56 | 76.1 | 77.4 | |
| FedProxParadigm=FL, Backbone=ResNet-502024.03 | 67.37 | 62.34 | 52 | 76.62 | 78.56 | |
| ResMLP-SParams.=40M, Architecture=ViT-based or MLP-like models2024.10 | 66.74 | 62.42 | 51.94 | 75.4 | 77.21 | |
| FedAvgParadigm=FL, Backbone=ResNet-502024.03 | 66.69 | 61.84 | 51.15 | 76.41 | 77.39 | |
| FedADGParadigm=FL+DG, Backbone=ResNet-502024.03 | 66.55 | 62.75 | 51.43 | 74.07 | 77.98 | |
| DeepAllParams.=23M, Architecture=CNN: ResNet-502024.10 | 66.5 | 61.3 | 52.4 | 75.8 | 76.6 | |
| MMDBackbone=ResNet-502023.03 | 66.4 | 60.4 | 53.3 | 74.3 | 77.4 | |
| DDAIG2022.05 | 65.5 | 59.2 | 52.3 | 74.6 | 76 | |
| MixStyle2022.05 | 65.5 | 58.7 | 53.4 | 74.2 | 75.9 | |
| RSCBackbone=ResNet-502023.03 | 65.5 | 60.7 | 51.4 | 74.8 | 75.1 | |
| ARMBackbone=ResNet-502023.03 | 64.8 | 58.9 | 51 | 74.1 | 75.2 | |
| FedSRParadigm=FL+DG, Backbone=ResNet-502024.03 | 64.73 | 62.75 | 49.85 | 72.24 | 74.1 | |
| IRMBackbone=ResNet-502023.03 | 64.3 | 58.9 | 52.2 | 72.1 | 74 | |
| RSC2022.05 | 63.12 | 58.42 | 47.9 | 71.63 | 74.54 | |
| L2A-OT2022.05 | 62.6 | 60.6 | 50.1 | 74.8 | 77 | |
| JiGen2022.05 | 61.2 | 53.04 | 47.51 | 71.47 | 72.79 | |
| MixstyleBackbone=ResNet-502023.03 | 61.1 | 51.1 | 53.2 | 68.2 | 72.1 | |
| D-SAM2022.05 | 60.77 | 58.03 | 44.37 | 69.22 | 71.45 | |
| PEERBackbone=ResNet-182025.05 | 60.63 | 56.81 | 54.23 | 70.84 | — | |
| PDENBackbone=ResNet-182025.05 | 54.34 | 53.39 | 43.38 | 66.25 | — | |
| L2DBackbone=ResNet-182025.05 | 54.03 | 54.02 | 41.77 | 66.3 | — | |
| ERMBackbone=ResNet-182025.05 | 53.9 | 52.78 | 40.19 | 68.73 | — | |
| M-ADABackbone=ResNet-182025.05 | 53.29 | 54.36 | 40.41 | 65.11 | — | |
| RandAugBackbone=ResNet-182025.05 | 50.01 | 43.1 | 45.47 | 61.67 | — |