Multi-modal Retrieval (Image Query) on WebQA
43.55Recall@20CLIP (ViT-L/14)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CLIP (ViT-L/14)Scoring Method=Std2025.11 | 43.55 | 22.54 | 25.76 | |
| Long-CLIP-LScoring Method=Std2025.11 | 39.84 | 20.22 | 23.36 | |
| CLIP (ViT-L/14)Scoring Method=Ours2025.11 | 37.1 | 20.37 | 22.75 | |
| Long-CLIP-LScoring Method=Ours2025.11 | 35.34 | 18.38 | 21.09 | |
| Long-CLIP-BScoring Method=Std2025.11 | 33.01 | 14.92 | 17.98 | |
| CLIP (ViT-B/32)Scoring Method=Std2025.11 | 32.82 | 15.2 | 18.02 | |
| Long-CLIP-BScoring Method=Ours2025.11 | 28.59 | 13.43 | 15.97 | |
| CLIP (ViT-B/32)Scoring Method=Ours2025.11 | 28.14 | 13.79 | 15.98 | |
| Cohere Embed 3Scoring Method=Std2025.11 | 27.42 | 12.36 | 14.83 | |
| Cohere Embed 3Scoring Method=Ours2025.11 | 25.39 | 11.48 | 13.73 | |
| E5-VScoring Method=Std2025.11 | 21.05 | 9.75 | 11.5 | |
| Cohere Embed 3Scoring Method=Cos2025.11 | 20.43 | 8 | 10.16 | |
| E5-VScoring Method=Ours2025.11 | 18.78 | 8.87 | 8.87 | |
| BLIPScoring Method=Std2025.11 | 17.62 | 8.24 | 9.73 | |
| BLIPScoring Method=Ours2025.11 | 14.04 | 6.35 | 7.62 | |
| E5-VScoring Method=Cos2025.11 | 11.89 | 5.19 | 6.37 | |
| CLIP (ViT-L/14)Scoring Method=Cos2025.11 | 0.04 | 0 | 0.01 | |
| CLIP (ViT-B/32)Scoring Method=Cos2025.11 | 0 | 0 | 0 | |
| Long-CLIP-BScoring Method=Cos2025.11 | 0 | 0 | 0 | |
| Long-CLIP-LScoring Method=Cos2025.11 | 0 | 0 | 0 | |
| BLIPScoring Method=Cos2025.11 | 0 | 0 | 0 |