Fine-grained concept removal on ImageNet Dogs
95.2Concept Fidelity (Df) Coarse TestME+GD
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ME+GD2025.06 | 95.2 | 53.2 | 45.12 | 46.79 | 43.04 | 88.69 | 57.52 | |
| HGA+KL2025.06 | 88.2 | 2 | 48.23 | 54.56 | 97.86 | 92.68 | 95.2 | |
| NPO+KL2025.06 | 88 | 8 | 49.33 | 53.91 | 91.43 | 93.06 | 92.24 | |
| Origin CLIP2025.06 | 86.2 | 93.4 | 50.88 | 65.55 | — | — | — | |
| Task vector2025.06 | 79.6 | 36.6 | 44.58 | 62.38 | 60.81 | 91.71 | 73.13 | |
| GA+KL2025.06 | 77.4 | 3 | 41.28 | 35.96 | 96.79 | 75.26 | 84.68 | |
| GDiff2025.06 | 69.6 | 0 | 40.54 | 9.3 | 100 | 58.21 | 73.58 | |
| SalUn2025.06 | 47.8 | 34.8 | 30.49 | 46.52 | 62.74 | 62.12 | 62.43 | |
| Relabeling2025.06 | 44.8 | 43.8 | 29.57 | 45.64 | 53.1 | 59.91 | 56.3 | |
| GA2025.06 | 1 | 0 | 7.8 | 1.57 | 100 | 6.3 | 11.85 |