Zero-shot Text-to-Image Retrieval on MS-COCO
7.17R@1BLIP3-o
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BLIP3-oEmbedding=POOLED-PATCH2026.05 | 7.17 | 37.49 | 70.5 | |
| GAP3DTraining=PRE-TRAINED, Embedding=POOLED-PATCH2026.05 | 4.98 | 24.02 | 44.98 | |
| GAP3DTraining=PRE-TRAINED, Embedding=CLS2026.05 | 4.47 | 23.06 | 44.18 | |
| GAP3DTraining=FINE-TUNED, Embedding=CLS2026.05 | 1.94 | 10.44 | 19.73 | |
| GAP3DTraining=FINE-TUNED, Embedding=POOLED-PATCH2026.05 | 0.27 | 1.77 | 4.08 |