Zero-shot text-to-image retrieval on Toys4K
7.42Recall@1BLIP3-o
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BLIP3-oEmbedding=POOLED-PATCH2026.05 | 7.42 | 42.52 | 80.06 | |
| GAP3DTraining=FINE-TUNED, Embedding=CLS2026.05 | 6.64 | 33.55 | 63.43 | |
| GAP3DTraining=PRE-TRAINED, Embedding=CLS2026.05 | 5.53 | 28.65 | 53.62 | |
| GAP3DTraining=FINE-TUNED, Embedding=POOLED-PATCH2026.05 | 2.96 | 14.81 | 27.48 | |
| GAP3DTraining=PRE-TRAINED, Embedding=POOLED-PATCH2026.05 | 1.54 | 9.09 | 16.86 |