Image Classification on Terra Incognita (TerraInc)
56.4AccuracyIB-ERM
Evaluation results
| Method | Links | |
|---|---|---|
| IB-ERM2021.06 | 56.4 | |
| IRM2021.06 | 54.6 | |
| IB-IRM2021.06 | 54.1 | |
| EVIL-SAMOptimization=SAM2025.12 | 50.5 | |
| MIROBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 50.4 | |
| SWADBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 50 | |
| ERM2021.06 | 49.8 | |
| SagNet-SAMOptimization=SAM2025.12 | 49.3 | |
| SagNetsBackbone=ResNet-502021.10 | 48.6 | |
| GMOE-S/16Backbone=ViT-S/16, Selection Criterion=train-validation selection2022.06 | 48.5 | |
| SparseIRM-SAMOptimization=SAM2025.12 | 48.5 | |
| mDSDIBackbone=ResNet-502021.10 | 48.1 | |
| MixupBackbone=ResNet-502021.10 | 47.9 | |
| CORAL-SAMOptimization=SAM2025.12 | 47.9 | |
| MLDGBackbone=ResNet-502021.10 | 47.7 | |
| IRMBackbone=ResNet-502021.10 | 47.6 | |
| CORALBackbone=ResNet-502021.10 | 47.6 | |
| IRMBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 47.6 | |
| FishrBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 47.4 | |
| MRM-SAMOptimization=SAM2025.12 | 47.4 | |
| ERMBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 47.2 | |
| DANNBackbone=ResNet-502021.10 | 46.7 | |
| RSCBackbone=ResNet-502021.10 | 46.6 | |
| VREXBackbone=ResNet-502021.10 | 46.4 | |
| DANNBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 46.4 | |
| CORALBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 46.4 | |
| ERMBackbone=ResNet-502021.10 | 46.1 | |
| CDANNBackbone=ResNet-502021.10 | 45.8 | |
| MTLBackbone=ResNet-502021.10 | 45.6 | |
| ARMBackbone=ResNet-502021.10 | 45.5 | |
| FISHBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 45.1 | |
| GroupDROBackbone=ResNet-502021.10 | 43.2 | |
| MMDBackbone=ResNet-502021.10 | 42.2 | |
| MMDBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 42.2 | |
| ERMBackbone=ViT-S/16, Selection Criterion=train-validation selection2022.06 | 42 | |
| FixMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 33.2 | |
| StyleMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 30.4 | |
| FixMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 28.9 | |
| FlexMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 28.8 | |
| StyleMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 28.7 | |
| FreeMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 26.5 | |
| FlexMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 24.9 | |
| FreeMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 23.5 | |
| ERMLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 22.9 | |
| EntMinLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 21.4 | |
| MeanTeacherLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 20.9 |