Domain Generalization on DomainNet
81.62AccuracyTask Experts
Evaluation Results
| Method | Links | |
|---|---|---|
| Task ExpertsBackbone=ViT-L-142026.03 | 81.62 | |
| Text-Anchored Information BottleneckBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=T2026.07 | 75.8 | |
| Task ExpertsBackbone=ViT-B-322026.03 | 74.43 | |
| Text-Anchored Information BottleneckBackbone=ViT-B/16, Pre-training=ImageNet-1k, Guidance (G)=T2026.07 | 68.8 | |
| PCBBackbone=ViT-L-142026.03 | 68.26 | |
| SCOREBackbone=ViT-L-142026.03 | 68.26 | |
| Iso-CTSBackbone=ViT-L-142026.03 | 67.63 | |
| Iso-CBackbone=ViT-L-142026.03 | 66.89 | |
| TSVBackbone=ViT-L-142026.03 | 66.79 | |
| ZeroshotBackbone=ViT-L-142026.03 | 66.25 | |
| TIESBackbone=ViT-L-142026.03 | 65.8 | |
| CLIPCEILBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 63.6 | |
| CLIPoodBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 63.5 | |
| CLIP-DTPBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 63.1 | |
| VL2VBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 62.8 | |
| Ours (PADG)Backbone=ViT-B/162025.07 | 62.74 | |
| CAR-FTBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 62.5 | |
| MagMaxBackbone=ViT-L-142026.03 | 62.24 | |
| CLIP-DPRBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 62.1 | |
| Ours (DPR)Backbone=ViT-B/16, Venue=CVPR’242025.07 | 62.05 | |
| SIMPLE+Backbone=ViT-B/16, Venue=ICLR’232025.07 | 61.9 | |
| MetaPromptFew-shot=true, Shots=5, CLIP-based=true2022.12 | 61.8 | |
| MetaPromptFew-shot=true, Shots=1, CLIP-based=true2022.12 | 61.5 | |
| CoCoOpFew-shot=true, Shots=5, CLIP-based=true2022.12 | 61.1 | |
| EoAEnsemble=true, CLIP-based=true2022.12 | 60.9 | |
| SPGBackbone=CLIP ViT-B/162024.04 | 60.1 | |
| SPGBackbone=ViT-B/16, Venue=ECCV’242025.07 | 60.1 | |
| CoOpBackbone=ViT-B/16, Venue=IJCV’222025.07 | 59.8 | |
| DUPRGBackbone=ViT-B/16, Venue=ICLR’232025.07 | 59.6 | |
| CoOpBackbone=CLIP ViT-B/162024.04 | 59.5 | |
| DPLBackbone=CLIP ViT-B/162024.04 | 59.5 | |
| MaPLeBackbone=CLIP ViT-B/162024.04 | 59.5 | |
| MaPLeBackbone=ViT-B/16, Venue=CVPR’232025.07 | 59.5 | |
| CoCoOpBackbone=CLIP ViT-B/162024.04 | 59.4 | |
| VL2VBackbone=ViT-B/16, Pre-training=ImageNet-1k, Guidance (G)=V,T2026.07 | 59.4 | |
| GESTURBackbone=ViT-B/16, Venue=ICCV’232025.07 | 58.9 | |
| GESTURBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=-2026.07 | 58.9 | |
| VL2V-SDBackbone=ViT-B/16, Venue=CVPR’242025.07 | 58.71 | |
| SCOREBackbone=ViT-B-322026.03 | 58.68 | |
| PCBBackbone=ViT-B-322026.03 | 58.53 | |
| VPTBackbone=CLIP ViT-B/162024.04 | 58.5 | |
| CLIP-LoRABackbone=ViT-B/16, Venue=CVPR’242025.07 | 58.4 | |
| Text-Anchored Information BottleneckBackbone=ResNet-50, Pre-training=ImageNet-1k, Guidance (G)=T2026.07 | 58.3 | |
| LPBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=-2026.07 | 58.2 | |
| CLIPZero-shot=true, CLIP-based=true2022.12 | 57.9 | |
| CLIP (template)Zero-shot=true, CLIP-based=true2022.12 | 57.8 | |
| CLIPBackbone=ViT-B/162025.07 | 57.5 | |
| Iso-CTSBackbone=ViT-B-322026.03 | 56.8 | |
| DPLBackbone=ViT-B/162025.07 | 56.7 | |
| ZS-CLIPBackbone=CLIP ViT-B/16, Evaluation Protocol=Zero-shot2024.04 | 56.6 | |
| RISE*Backbone=ViT-B/16, Pre-training=ImageNet-1k, Guidance (G)=V,T2026.07 | 56.6 | |
| VPBackbone=CLIP ViT-B/162024.04 | 56.5 | |
| ZeroshotBackbone=ViT-B-322026.03 | 56.1 | |
| SMABackbone=ViT-B/16, Venue=NIPS’222025.07 | 55.9 | |
| Iso-CBackbone=ViT-B-322026.03 | 55.84 | |
| TSVBackbone=ViT-B-322026.03 | 55.64 | |
| RISEBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=V,T2026.07 | 55.4 | |
| SEDGEEnsemble=true, CLIP-based=true2022.12 | 54.7 | |
| SEDGEBackbone=ViT-B/162025.07 | 54.7 | |
| MIRO (ViT-B/16)Backbone=ViT-B/16, Venue=ECCV’222025.07 | 54 | |
| MIROBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=-2026.07 | 54 | |
| ERM (ViT-B/16)Backbone=ViT-B/16, Venue=ICLR’212025.07 | 53.8 | |
| TIESBackbone=ViT-B-322026.03 | 53.22 | |
| SWADBackbone=ViT-B/16, Venue=NIPS’212025.07 | 51.7 | |
| CADGEnsemble status=Non-ensemble2022.03 | 51.6 | |
| LPBackbone=ViT-B/16, Pre-training=ImageNet-1k, Guidance (G)=-2026.07 | 50.5 | |
| PAPTVenue=-2026.06 | 50.18 | |
| SPGBackbone=CLIP RN502024.04 | 50.1 | |
| CoOpBackbone=CLIP RN502024.04 | 49.7 | |
| CoCoOpBackbone=CLIP RN502024.04 | 49.7 | |
| CLIP-ZSBackbone=CLIP-ViT-B/16, Pre-training=private dataset (400M), Guidance (G)=-2026.07 | 49.7 | |
| DPLBackbone=CLIP RN502024.04 | 49.6 | |
| GMoE-S/16Backbone=GMoE-S/16, Selection Criterion=leave-one-domain-out2022.06 | 48.4 | |
| Lin. Prob.Backbone=CLIP ViT-B/16, Evaluation Protocol=Linear Probing2024.04 | 48.2 | |
| INSUREBackbone=ResNet-50, Pre-training=ImageNet-1k, Guidance (G)=-2026.07 | 48 | |
| VL2VBackbone=ResNet-50, Pre-training=ImageNet-1k, Guidance (G)=V,T2026.07 | 47.7 | |
| SWAD + POEMSetting=DomainBed settings (reproduced)2023.05 | 47.1 | |
| MIRO + SWAD + POEMSetting=DomainBed settings (reproduced)2023.05 | 47.1 | |
| MIRO + SWADSetting=DomainBed settings (reproduced)2023.05 | 46.9 | |
| Previous SOTA [31] + SWAD2021.02 | 46.8 | |
| RESBackbone=ResNet-50, Venue=ECCV’242025.07 | 46.7 | |
| RESVenue=ECCV’242026.06 | 46.7 | |
| SWAD2021.02 | 46.5 | |
| SWADEnsemble=true2022.12 | 46.5 | |
| SWADSetting=Reported2023.05 | 46.5 | |
| SWADBackbone=RN502024.04 | 46.5 | |
| RISEBackbone=ResNet-50, Pre-training=ImageNet-1k, Guidance (G)=V,T2026.07 | 46.5 | |
| ZS-CLIPBackbone=CLIP RN50, Evaluation Protocol=Zero-shot2024.04 | 46.4 | |
| GESTURBackbone=ResNet-50, Pre-training=ImageNet-1k, Guidance (G)=-2026.07 | 46.3 | |
| SWADSetting=DomainBed settings (reproduced)2023.05 | 46.2 | |
| ViT-S/16Backbone=ViT-S/16, Selection Criterion=leave-one-domain-out2022.06 | 46.1 | |
| VPBackbone=CLIP RN502024.04 | 45.8 | |
| DSP-Reg2026.01 | 45.6 | |
| SMOSBackbone=ResNet-50, Venue=CVPR’242025.07 | 45.3 | |
| SMOSVenue=CVPR’242026.06 | 45.3 | |
| GGA2026.01 | 45.2 | |
| FGMix + SWAModel Selection Strategy=training domain validation, Optimizer=SWA, Training Time (ms/iter)=319.82022.09 | 45.1 | |
| SAGMGGA=Disabled2025.02 | 45 | |
| SAGM2026.01 | 45 | |
| SAGMBackbone=ResNet-50, Venue=CVPR’232025.07 | 45 |