Image Classification on Waterbirds 100% correlation (test)
91.9Worst-group AccuracyBAP (LP)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BAP (LP)Backbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Linear Probing2026.05 | 91.9 | 94 | |
| Data-Matched ControlBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 80.7 | 87.2 | |
| SupERGroup Info=false, Train Twice=false2025.08 | 79.7 | 85 | |
| BAP (ZS)Backbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 71.6 | 86.4 | |
| DIALBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 64.9 | 79.7 | |
| CnCGroup Info=false, Train Twice=true2025.08 | 62.1 | 81.9 | |
| LfFGroup Info=false, Train Twice=false2025.08 | 61.5 | 80.6 | |
| JTTGroup Info=false, Train Twice=true2025.08 | 61.3 | 79.7 | |
| RoboShotBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 59.2 | 75.8 | |
| CVaR DROGroup Info=false, Train Twice=false2025.08 | 58 | 79 | |
| IRMGroup Info=true, Train Twice=false2025.08 | 57 | 80.5 | |
| GroupDROGroup Info=true, Train Twice=false2025.08 | 56.5 | 79.4 | |
| UWGroup Info=true, Train Twice=false2025.08 | 56.4 | 78.6 | |
| GALSGroup Info=false, Train Twice=false2025.08 | 55 | 79.7 | |
| DFRGroup Info=true, Train Twice=true2025.08 | 48.2 | 76.4 | |
| Native BackboneBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 47.2 | 75.3 | |
| ERMGroup Info=false, Train Twice=false2025.08 | 46.4 | 74.8 | |
| PruSCBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 35.6 | 47.3 | |
| Native Backbone (LP)Backbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Linear Probing2026.05 | 31.3 | 63.3 | |
| WiSE-FTBackbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 21.6 | 63.1 | |
| ERM (LP-FT)Backbone=ConvNeXT-W, Pre-training=CLIP, Evaluation Protocol=Linear Probing & Fine-Tuning2026.05 | 18.7 | 63.1 |