Image Classification on PACS
98.47AccuracyAny-shift Prompting + TPT
Evaluation results
| Method | Links | |
|---|---|---|
| Any-shift Prompting + TPTTest-time optimization=true, Backbone=ViT-B/162024.02 | 98.47 | |
| Any-shift PromptingTest-time optimization=false, Backbone=ViT-B/162024.02 | 98.16 | |
| CoCoOp + TPTTest-time optimization=true, Backbone=ViT-B/162024.02 | 97.95 | |
| CoOp + TPTTest-time optimization=true, Backbone=ViT-B/162024.02 | 97.85 | |
| TPTTest-time optimization=true, Backbone=ViT-B/162024.02 | 97.25 | |
| DPLTest-time optimization=false, Backbone=ViT-B/162024.02 | 97.07 | |
| CoCoOpTest-time optimization=false, Backbone=ViT-B/162024.02 | 97 | |
| CLIP-DTest-time optimization=false, Backbone=ViT-B/162024.02 | 96.65 | |
| SGD-DualOpt_Fine-tuneBackbone=ConvNeXt-B2026.04 | 96.63 | |
| CoOpTest-time optimization=false, Backbone=ViT-B/162024.02 | 96.45 | |
| CLIPTest-time optimization=false, Backbone=ViT-B/162024.02 | 96.13 | |
| L2-SPBackbone=ConvNeXt-B2026.04 | 95.9 | |
| FullBackbone=ConvNeXt-B2026.04 | 95.51 | |
| Adam-DualOpt_Fine-tuneBackbone=ViT-B2026.04 | 94.38 | |
| VPTBackbone=ConvNeXt-B2026.04 | 93.75 | |
| LoRABackbone=ConvNeXt-B2026.04 | 93.7 | |
| LinearBackbone=ConvNeXt-B2026.04 | 93.46 | |
| CLAdapterEvaluation Protocol=Fine-Tuning, Backbone=ViT-B2025.12 | 91.41 | |
| CLAdapterEvaluation Protocol=Parameter-Efficient Fine-Tuning, Backbone=ViT-B2025.12 | 91.41 | |
| LoRABackbone=ViT-B2026.04 | 90.48 | |
| TSDBackbone=ResNet-502023.03 | 89.41 | |
| MoRAEvaluation Protocol=Parameter-Efficient Fine-Tuning, Backbone=ViT-B2025.12 | 89.09 | |
| SFTEvaluation Protocol=Fine-Tuning, Backbone=ViT-B2025.12 | 88.91 | |
| LoRAEvaluation Protocol=Parameter-Efficient Fine-Tuning, Backbone=ViT-B2025.12 | 88.53 | |
| DoRAEvaluation Protocol=Parameter-Efficient Fine-Tuning, Backbone=ViT-B2025.12 | 88.43 | |
| SWADBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 88.1 | |
| GMOE-S/16Backbone=ViT-S/16, Selection Criterion=train-validation selection2022.06 | 88.1 | |
| EVIL-SAMOptimization=SAM2025.12 | 87.8 | |
| FullEvaluation Protocol=Fine-Tuning, Backbone=ViT-B2025.12 | 87.79 | |
| FullBackbone=ViT-B2026.04 | 87.79 | |
| L2-SPEvaluation Protocol=Fine-Tuning, Backbone=ViT-B2025.12 | 87.74 | |
| L2-SPBackbone=ViT-B2026.04 | 87.74 | |
| TentBackbone=ResNet-502023.03 | 87.48 | |
| TSDBackbone=ResNet-182023.03 | 87.32 | |
| LAMEBackbone=ResNet-502023.03 | 86.62 | |
| CORAL-SAMOptimization=SAM2025.12 | 86.6 | |
| SagNet-SAMOptimization=SAM2025.12 | 86.4 | |
| SagNetsBackbone=ResNet-502021.10 | 86.3 | |
| CORALBackbone=ResNet-502021.10 | 86.2 | |
| mDSDIBackbone=ResNet-502021.10 | 86.2 | |
| ERMBackbone=ViT-S/16, Selection Criterion=train-validation selection2022.06 | 86.2 | |
| T3ABackbone=ResNet-502023.03 | 86.04 | |
| CORALBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 86 | |
| ERMBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 85.7 | |
| ERMBackbone=ResNet-502021.10 | 85.5 | |
| FISHBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 85.5 | |
| FishrBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 85.5 | |
| SHOT-IMBackbone=ResNet-502023.03 | 85.5 | |
| MIROBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 85.4 | |
| PLBackbone=ResNet-502023.03 | 85.23 | |
| RSCBackbone=ResNet-502021.10 | 85.2 | |
| SparseIRM-SAMOptimization=SAM2025.12 | 85.2 | |
| ARMBackbone=ResNet-502021.10 | 85.1 | |
| ETABackbone=ResNet-502023.03 | 85.04 | |
| BNBackbone=ResNet-502023.03 | 85.03 | |
| MMDBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 85 | |
| TentBackbone=ResNet-182023.03 | 84.92 | |
| MLDGBackbone=ResNet-502021.10 | 84.9 | |
| VREXBackbone=ResNet-502021.10 | 84.9 | |
| PLBackbone=ResNet-182023.03 | 84.64 | |
| MixupBackbone=ResNet-502021.10 | 84.6 | |
| MMDBackbone=ResNet-502021.10 | 84.6 | |
| MTLBackbone=ResNet-502021.10 | 84.6 | |
| DANNBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 84.6 | |
| ERMBackbone=ResNet-502023.03 | 84.59 | |
| LAMEBackbone=ResNet-182023.03 | 84.58 | |
| GroupDROBackbone=ResNet-502021.10 | 84.4 | |
| MRM-SAMOptimization=SAM2025.12 | 83.9 | |
| DANNBackbone=ResNet-502021.10 | 83.6 | |
| IRMBackbone=ResNet-502021.10 | 83.5 | |
| IRMBackbone=ResNet-50, Selection Criterion=train-validation selection2022.06 | 83.5 | |
| T3ABackbone=ResNet-182023.03 | 83.5 | |
| BNBackbone=ResNet-182023.03 | 82.82 | |
| ETABackbone=ResNet-182023.03 | 82.7 | |
| CDANNBackbone=ResNet-502021.10 | 82.6 | |
| SHOT-IMBackbone=ResNet-182023.03 | 82.55 | |
| ERMBackbone=ResNet-182023.03 | 82.07 | |
| StyleMatchLabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 79.4 | |
| StyleMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 79.3 | |
| Feature ModulationLabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 78.7 | |
| StyleMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 78.4 | |
| StyleMatchLabels per class=5, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 78.4 | |
| DGWMLabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 78.4 | |
| FBCSALabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 78.2 | |
| DGWMLabels per class=5, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 77.9 | |
| VPTBackbone=ViT-B2026.04 | 77.44 | |
| FixMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 77.3 | |
| FBCSALabels per class=5, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 77.3 | |
| VPT-AdapterEvaluation Protocol=Parameter-Efficient Fine-Tuning, Backbone=ViT-B2025.12 | 76.76 | |
| FixMatchLabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 76.6 | |
| DualTTA2026.04 | 76.02 | |
| Feature ModulationLabels per class=5, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 75.4 | |
| DeYO2026.04 | 75.16 | |
| Tent2026.04 | 74.96 | |
| FreeMatch + OursLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 73.7 | |
| FixMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 73.4 | |
| FixMatchLabels per class=5, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 73.4 | |
| SAR2026.04 | 73.26 | |
| FreeMatchLabels per class=5, Evaluation protocol=leave-one-domain out, Average over seeds=52024.03 | 72.8 | |
| FlexMatchLabels per class=10, Backbone=ResNet-18, Pre-trained=ImageNet, Evaluation protocol=leave-one-domain-out2025.03 | 72.7 |