Multi-modal Retrieval (ALL->T) on EVQA+
33.36R@1MiMIC-ANCE
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MiMIC-ANCETraining Strategy=ANCE2026.04 | 33.36 | 47.9 | 61.28 | 73.87 | 44.21 | 43.64 | 44.51 | 45.28 | |
| Marvel-ANCETraining Strategy=ANCE2026.04 | 31.85 | 45.26 | 57.51 | 66.8 | 42.26 | 41.5 | 42.5 | 42.96 | |
| UniVL-DR2026.04 | 30.48 | 45.35 | 58.9 | 70.75 | 41.58 | 41.02 | 41.89 | 42.66 | |
| MiMICANCE Training=False2026.04 | 29.41 | 43.08 | 57.06 | 71.13 | 40.03 | 39.3 | 40.36 | 41.15 | |
| CLIP-DPR2026.04 | 27.86 | 41.19 | 55.32 | 68.34 | 38.38 | 37.62 | 38.72 | 39.39 | |
| Marvel-DPR2026.04 | 26.17 | 37.92 | 49.75 | 59.55 | 35.74 | 34.74 | 36.02 | 36.3 | |
| VISTA*Backbone=CLIP-B-322026.04 | 24.05 | 36.16 | 49.46 | 60.27 | 33.78 | 33.05 | 34.05 | 34.73 |