List your company
ResearchFormerly WizWand

Domain Generalization on TerraIncognita (test)

78.4AccuracyText-Anchored Information Bottleneck
15.16831.5844864.416Nov 25, 2021Sep 1, 2022Jun 8, 2023Mar 14, 2024Dec 19, 2024Sep 25, 2025Jul 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Text-Anchored Information BottleneckBackbone=Swin Transformer, Pre-training source=IN-21K2026.0778.4
Text-Anchored Information BottleneckBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0776.7
Text-Anchored Information BottleneckBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0773.5
Text-Anchored Information BottleneckBackbone=DeiT, Pre-training source=IN-1K2026.0771.5
Text-Anchored Information BottleneckBackbone=DINOv2, Pre-training source=LVD-142M2026.0763.6
CLIPoodBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0762.5
VL2VBackbone=DINOv2, Pre-training source=LVD-142M2026.0761.6
VL2VBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0761.1
MIROBackbone=DINOv2, Pre-training source=LVD-142M2026.0760.4
RISEBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0760.2
MIROBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0758.9
CLIPoodBackbone=DINOv2, Pre-training source=LVD-142M2026.0758.1
MIROBackbone=Swin Transformer, Pre-training source=IN-21K2026.0757.7
CLIPoodBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0757.5
LPBackbone=DINOv2, Pre-training source=LVD-142M2026.0757.4
VL2VBackbone=Swin Transformer, Pre-training source=IN-21K2026.0757
LPBackbone=Swin Transformer, Pre-training source=IN-21K2026.0755.5
VL2VBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0755.3
LPBackbone=RegNetY-16GF, Pre-training source=IG-3B2026.0755.2
RISE (ERM + Hint + AD + RD + MT)Backbone=ResNet50, Ens/MA=Yes2023.0954
DAMPEvaluation Protocol=Prompt Learning, Backbone=ViT-B/162024.0353.7
CORALEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.0353.5
RISEBackbone=Swin Transformer, Pre-training source=IN-21K2026.0753.4
VL2VBackbone=DeiT, Pre-training source=IN-1K2026.0753.4
DPLEvaluation Protocol=Prompt Learning, Backbone=ViT-B/162024.0352.6
EoABackbone=ResNet50, Ens/MA=Yes2023.0952.3
RISE (ERM + Hint + AD + RD)Backbone=ResNet50, Ens/MA=No2023.0952.3
ERM + Hint + ADBackbone=ResNet50, Ens/MA=No2023.0952.2
DANNEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.0352.2
GENIEBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0652
ERM + Hint + RDBackbone=ResNet50, Ens/MA=No2023.0951.7
ERM + Pseudo LabelBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label2021.1150.5
ERM + HintBackbone=ResNet50, Ens/MA=No2023.0950.5
MIROBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0750.4
CLIP (Teacher)Backbone=ViT B/16, Ens/MA=No, Fine-tuning=true2023.0950.2
ERMEvaluation Protocol=Fine-tuning (CLIP), Backbone=ViT-B/162024.0350.2
SWADBackbone=ResNet50, Ens/MA=Yes2023.0950
MIROBackbone=DeiT, Pre-training source=IN-1K2026.0749.9
CLIPoodBackbone=DeiT, Pre-training source=IN-1K2026.0749.8
LPBackbone=DeiT, Pre-training source=IN-1K2026.0749.5
CLIP + DPLBackbone=CLIP ViT-B16, Test-time batch size=642021.1149.4
EnsembleBackbone=ResNet50, Ens/MA=Yes2023.0949.2
CORAL + T3ABackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=T3A2021.1149
CORAL + Tent*Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Tent, Update target=linear classifier2021.1148.7
CORALBackbone=CLIP ViT-B16, Test-time batch size=642021.1148.6
CORAL + Pseudo Label*Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label, Update target=linear classifier2021.1148.6
Best SOTA competitorBackbone=ResNet50, Ens/MA=No2023.0948.6
RISE (ERM + Hint + AD + RD)Backbone=ResNet18, Ens/MA=No2023.0948.4
ERM + T3ABackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=T3A2021.1148.2
ERM + Hint + ADBackbone=ResNet18, Ens/MA=No2023.0948.2
ERM + Hint + RDBackbone=ResNet18, Ens/MA=No2023.0947.9
ERMBackbone=CLIP ViT-B16, Test-time batch size=642021.1147.8
ERMBackbone=ResNet50, Ens/MA=No2023.0947.2
VL2VBackbone=EfficientNet, Pre-training source=IN-1K2026.0747.2
ERM + HintBackbone=ResNet18, Ens/MA=No2023.0947
ERM + Pseudo Label*Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label, Update target=linear classifier2021.1146.9
RISEBackbone=DeiT, Pre-training source=IN-1K2026.0746.9
Text-Anchored Information BottleneckBackbone=EfficientNet, Pre-training source=IN-1K2026.0746.8
ERM + Tent*Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Tent, Update target=linear classifier2021.1146.7
LPBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0746.3
SAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.7
FADBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.7
YOGIBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.4
ADAMWBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.2
ADABELIEFBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.2
GAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0645.2
RISEBackbone=EfficientNet, Pre-training source=IN-1K2026.0745.1
SGDBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0644.9
SWAD* + BatchFormerBackbone=ResNet-182022.0344.8
CORAL + Pseudo LabelBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=Pseudo Label2021.1144.5
Best SoTA competitorBackbone=ResNet18, Ens/MA=No2023.0944.5
ADAMBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0644.4
ADAHESSIANBackbone=ResNet-50, Pre-training=ImageNet, Training Iterations=5,000, Evaluation Protocol=DomainBed (Training-domain Validation Set)2026.0644.4
CLIPoodBackbone=EfficientNet, Pre-training source=IN-1K2026.0743.8
ERMBackbone=ResNet18, Ens/MA=No2023.0943.6
SWAD*Backbone=ResNet-182022.0342.1
CORAL + SHOTBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT2021.1141.9
CORAL + SHOT**Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT, Update target=feature extractor2021.1141.9
ERM + SHOT**Backbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT, Update target=feature extractor2021.1140.6
MIROBackbone=EfficientNet, Pre-training source=IN-1K2026.0740.6
ERM + SHOTBackbone=CLIP ViT-B16, Test-time batch size=64, Test-time adaptation method=SHOT2021.1140.5
CLIPoodBackbone=Swin Transformer, Pre-training source=IN-21K2026.0740.4
RISEBackbone=DINOv2, Pre-training source=LVD-142M2026.0740.2
LPBackbone=EfficientNet, Pre-training source=IN-1K2026.0739.6
CORAL+GENIEBackbone=ResNet-502026.0636.7
GENIEBackbone=ResNet-502026.0636
RISEBackbone=RegNetY-16GF, Pre-training source=IN-12K2026.0735.1
CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/162024.0334.1
ADAMBackbone=ResNet-502026.0633.5
CORAL+ADAMBackbone=ResNet-502026.0633.5
RSC+GENIEBackbone=ResNet-502026.0633.2
RSC+ADAMBackbone=ResNet-502026.0631.6
SAMBackbone=ResNet-502026.0626.8
SGDBackbone=ResNet-502026.0622.8
CORAL+SGDBackbone=ResNet-502026.0622.8
RSC+SGDBackbone=ResNet-502026.0617.6