Image Classification on CelebA (test) (WG/Avg/Gap)
92.3WG ScoreDKT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DKT2025.02 | 92.3 | 94.6 | — | |
| GroupDRO2025.02 | 88.9 | 92.9 | — | |
| DATBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 84.94 | 86.54 | 1.19 | |
| DAT*Backbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 84.93 | 86.54 | 1.61 | |
| TIEBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 84.6 | 86.17 | 1.57 | |
| EvA-E2025.02 | 82.7 | 88.7 | — | |
| TIEBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 82.63 | 85.11 | 2.48 | |
| TIE*Backbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 82.61 | 85.1 | 2.49 | |
| ROBOSHOTBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 82.61 | 85.54 | 2.93 | |
| TIE*Backbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 81.98 | 84.27 | 2.29 | |
| JTT2025.02 | 81.1 | 88 | — | |
| DATBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 80.79 | 87.09 | 6.3 | |
| ROBOSHOTBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 80.52 | 84.77 | 4.25 | |
| Perception CLIPBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 80.22 | 85.17 | 4.95 | |
| ZSBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 78.89 | 84.27 | 5.38 | |
| Perception CLIPBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 78.7 | 81.41 | 2.71 | |
| DATBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 78.53 | 87.09 | 8.56 | |
| DAT*Backbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 78.53 | 87.11 | 8.58 | |
| DAT*Backbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 78.53 | 88.29 | 9.76 | |
| Ideal wordsBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 78.12 | 80.96 | 2.84 | |
| Orth-CaliBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 77.92 | 82.31 | 4.39 | |
| Orth-CaliBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 77.69 | 81.39 | 3.7 | |
| Ideal wordsBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 76.67 | 89.15 | 12.48 | |
| Perception CLIPBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 76.46 | 80.32 | 3.86 | |
| TIEBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 75.32 | 81.71 | 6.39 | |
| TIE*Backbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 75.3 | 81.7 | 6.4 | |
| Group PromptBackbone=CLIP (ViT-B/32), Evaluation Protocol=Zero-shot2026.06 | 74.9 | 80.38 | 5.48 | |
| IPG2025.02 | 74.7 | 90.8 | — | |
| ROBOSHOTBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 73.96 | 80.9 | 6.94 | |
| ZSBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 73.35 | 81.2 | 7.85 | |
| LfF2025.02 | 70.6 | 86 | — | |
| Group PromptBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 70.59 | 79.48 | 8.89 | |
| ZSBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 69.69 | 81.58 | 11.89 | |
| Orth-CaliBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 69.13 | 76.47 | 7.34 | |
| Group PromptBackbone=CLIP (ViT-L/14), Evaluation Protocol=Zero-shot2026.06 | 68.94 | 77.86 | 8.92 | |
| Ideal wordsBackbone=CLIP (ResNet50), Evaluation Protocol=Zero-shot2026.06 | 65.65 | 76.27 | 10.62 | |
| ERM2025.02 | 47.8 | 94.9 | — |