Image-Text Retrieval Robustness on Urban1k
77.6Word Dropout (p=0.3)HyFL-CLIP
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HyFL-CLIP2026.07 | 77.6 | 77.14 | 75.45 | — | 86.08 | 27.64 | 36.25 | |
| HiMo-CLIPimplementation=evaluated using our own implementation2026.07 | 73.47 | 54.98 | 68.95 | — | 81.16 | 24.77 | 32.74 | |
| FineLIPimplementation=evaluated using our own implementation2026.07 | 73.45 | 54.19 | 68.2 | — | 82.57 | 24.7 | 31.96 | |
| Long-CLIPsource=obtained using checkpoints provided by the original authors2026.07 | 61.73 | 44.78 | 59.4 | — | 68.82 | 22.51 | 27.98 |