Domain Generalization on VLCS (test)
88.8AccuracyText-Anchored Information Bottleneck
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Text-Anchored Information BottleneckBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 88.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-Anchored Information BottleneckBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 88.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-Anchored Information BottleneckBackbone=DeiT, Pre-training source=IN-1K2026.07 | 88.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-Anchored Information BottleneckBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 87.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-Anchored Information BottleneckBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DAMPEvaluation Protocol=Prompt Learning, Backbone=ViT-B/162024.03 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DPLEvaluation Protocol=Prompt Learning, Backbone=ViT-B/162024.03 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-Anchored Information BottleneckBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 83.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=DeiT, Pre-training source=IN-1K2026.07 | 83.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DANNEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.03 | 83.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 83 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 82.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERMEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.03 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISE (ERM + Hint + AD + RD + MT)Backbone=ResNet50, Ens/MA=Yes2023.09 | 82.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 82.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 82.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP (template)Zero-shot=true, CLIP=true2022.12 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP (Teacher)Backbone=ViT B/16, Ens/MA=No, Inference Mode=Zero-shot2023.09 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/162024.03 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEDGEEnsemble=true, CLIP=true2022.12 | 82.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaPromptFew-shot=5-shot, CLIP=true2022.12 | 82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CORALEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.03 | 82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaPromptFew-shot=1-shot, CLIP=true2022.12 | 81.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISE (ERM + Hint + AD + RD)Backbone=ResNet50, Ens/MA=No2023.09 | 81.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 81.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 81.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + Hint + ADBackbone=ResNet50, Ens/MA=No2023.09 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=DeiT, Pre-training source=IN-1K2026.07 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=DINOv2, Pre-training source=LVD-142M2026.07 | 81.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + Hint + RDBackbone=ResNet50, Ens/MA=No2023.09 | 81.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EoAEnsemble=true, CLIP=true2022.12 | 81.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=DeiT, Pre-training source=IN-1K2026.07 | 81 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 81 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISEBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 80.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + HintBackbone=ResNet50, Ens/MA=No2023.09 | 80.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GENIEBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 80.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 80.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VL2VBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 80.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 80.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=DeiT, Pre-training source=IN-1K2026.07 | 80.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=DeiT, Pre-training source=IN-1K2026.07 | 79.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.07 | 79.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SWADEnsemble=true2022.12 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SWADBackbone=ResNet50, Ens/MA=Yes2023.09 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EoABackbone=ResNet50, Ens/MA=Yes2023.09 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=EfficientNet, Pre-training source=IN-1K2026.07 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIROBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 79 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FADBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL2022.12 | 78.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Best SOTA competitorBackbone=ResNet50, Ens/MA=No2023.09 | 78.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RISE (ERM + Hint + AD + RD)Backbone=ResNet18, Ens/MA=No2023.09 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADAHESSIANBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EnsembleBackbone=ResNet50, Ens/MA=Yes2023.09 | 78.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + Hint + ADBackbone=ResNet18, Ens/MA=No2023.09 | 78.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADABELIEFBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 78.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoCoOpFew-shot=5-shot, CLIP=true2022.12 | 78.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + Hint + RDBackbone=ResNet18, Ens/MA=No2023.09 | 78.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGDBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 78.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.07 | 78.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + HintBackbone=ResNet18, Ens/MA=No2023.09 | 78 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Fish2022.12 | 77.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| YOGIBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 77.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERMBackbone=ResNet50, Ens/MA=No2023.09 | 77.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADAMWBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 77.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM2022.12 | 77.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.06 | 77.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MLDG2022.12 | 77.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPoodBackbone=Swin Transformer, Pre-training source=IN-21K2026.07 | 76.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPZero-shot=true, CLIP=true2022.12 | 75.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Best SoTA competitorBackbone=ResNet18, Ens/MA=No2023.09 | 74.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERMBackbone=ResNet18, Ens/MA=No2023.09 | 73.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GENIEBackbone=ResNet-502026.06 | 69.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL+GENIEBackbone=ResNet-502026.06 | 69.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RSC+GENIEBackbone=ResNet-502026.06 | 68.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAMBackbone=ResNet-502026.06 | 66.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGDBackbone=ResNet-502026.06 | 60.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL+SGDBackbone=ResNet-502026.06 | 60.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADAMBackbone=ResNet-502026.06 | 56.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL+ADAMBackbone=ResNet-502026.06 | 56.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RSC+ADAMBackbone=ResNet-502026.06 | 51.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RSC+SGDBackbone=ResNet-502026.06 | 39.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Algorithm 1 SelectionSelection Criterion=Algorithm 12021.06 | — | 70.97 | 66.98 | 97.81 | 69.5 | 76.32 | 0.63 | — | — | — | — | — | — | — | — | — | — | |
| Baselinebackbone=AlexNet2020.07 | — | 70.58 | 59.72 | 96.25 | 64.51 | 72.76 | — | — | — | — | — | — | — | — | — | — | — | |
| CCSTBackbone=ResNet, Data Distribution=Single-domain, augmentation level=32023.11 | — | 68.2 | 61.2 | 92.5 | 66.5 | 72.1 | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP + DPLBackbone=CLIP ViT-B16, Test-time batch size=642021.11 | — | — | — | — | — | 81 | — | — | — | — | — | — | — | — | — | — | — | |
| CORALBackbone=CLIP ViT-B16, Test-time batch size=642021.11 | — | — | — | — | — | 81.2 | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL + Pseudo LabelBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label2021.11 | — | — | — | — | — | 80 | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL + Pseudo Label*Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label, Update target=linear classifier2021.11 | — | — | — | — | — | 81.4 | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL + SHOTBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT2021.11 | — | — | — | — | — | 78.7 | — | — | — | — | — | — | — | — | — | — | — | |
| CORAL + SHOT**Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT, Update target=feature extractor2021.11 | — | — | — | — | — | 78.5 | — | — | — | — | — | — | — | — | — | — | — |