Gender Classification on COCO 100% spurious correlation
77.9Average ScoreBAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BAPBackbone=CLIP, Evaluation Protocol=Linear Probing2026.05 | 77.9 | 74.7 | |
| BAPBackbone=SigLIP2, Evaluation Protocol=Linear Probing2026.05 | 77.8 | 75.3 | |
| BAPBackbone=SigLIP2, Evaluation Protocol=Zero-Shot2026.05 | 77.5 | 71.8 | |
| DIALBackbone=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 75.1 | 53.8 | |
| Data-Matched ControlBackbone=SigLIP2, Evaluation Protocol=Trained Evaluation2026.05 | 74.8 | 63.5 | |
| BAPBackbone=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 74.2 | 57.1 | |
| RoboShotBackbone=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 73.2 | 52.5 | |
| Data-Matched ControlBackbone=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 73.2 | 64.6 | |
| RoboShotBackbone=SigLIP2, Evaluation Protocol=Zero-Shot2026.05 | 72.9 | 67 | |
| DIALBackbone=SigLIP2, Evaluation Protocol=Zero-Shot2026.05 | 72.4 | 65.2 | |
| Native BackboneBackbone=CLIP, Evaluation Protocol=Zero-Shot2026.05 | 71.4 | 48.3 | |
| Native BackboneBackbone=SigLIP2, Evaluation Protocol=Zero-Shot2026.05 | 71.3 | 64.2 | |
| WiSE-FTBackbone=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 65.5 | 32.7 | |
| ERMBackbone=CLIP, Evaluation Protocol=Linear Probing & Fine-Tuning2026.05 | 63.1 | 4.9 | |
| ERMBackbone=SigLIP2, Evaluation Protocol=Linear Probing & Fine-Tuning2026.05 | 55.3 | 6.2 | |
| WiSE-FTBackbone=SigLIP2, Evaluation Protocol=Trained Evaluation2026.05 | 54.9 | 36.1 | |
| PruSCBackbone=SigLIP2, Evaluation Protocol=Trained Evaluation2026.05 | 54.6 | 13.7 | |
| PruSCBackbone=CLIP, Evaluation Protocol=Trained Evaluation2026.05 | 53 | 11.5 | |
| Native BackboneBackbone=SigLIP2, Evaluation Protocol=Linear Probing2026.05 | 51.4 | 9.3 | |
| Native BackboneBackbone=CLIP, Evaluation Protocol=Linear Probing2026.05 | 49.6 | 5.3 |