Entity Retrieval on E-VQA (val)
36.5Recall@1EchoSight
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| EchoSightType=Trained multimodal re-ranker2026.06 | 36.5 | 45.3 | 47.9 | 48.8 | 48.8 | |
| IBASelection=top-3 entities from 20 retrieved candidates, Tuning=Dataset-specific tuned weights2026.06 | 35.5 | 43.3 | — | — | — | |
| IBASelection=top-3 entities from 20 retrieved candidates, Tuning=No dataset-specific tuning (all score-fusion weights set to 1)2026.06 | 34.9 | — | — | — | — | |
| EVA-CLIPBackbone=EVA-CLIP-8B, Type=Initial image-to-image retrieval2026.06 | 13.4 | 26.1 | 31.9 | 41.8 | 48.8 |