Text-to-Image Retrieval on MS-COCO (Hard)
43.5Recall@1BEIT-3
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BEIT-3Config=COCO-Finetuned, Zero-Shot=false2026.02 | 43.5 | 73.9 | 87 | |
| Visual Model Checking + BEITZero-Shot=false2026.02 | 39.1 | 71.7 | 91.3 | |
| SigLIPParameters=400M, Zero-Shot=true2026.02 | 30.4 | 47.8 | 69.6 | |
| CLIPBackbone=ViT-H, QuickGELU=true, Zero-Shot=true2026.02 | 28.3 | 41.3 | 67.4 | |
| ALIGNZero-Shot=true2026.02 | 28.3 | 41.3 | 56.5 | |
| Visual Model Checking + ALIGNZero-Shot=true2026.02 | 26.1 | 54.3 | 71.7 | |
| Visual Model Checking (base)Zero-Shot=true2026.02 | 19.6 | 49 | 73.9 | |
| CLIPBackbone=ViT-B, Zero-Shot=true2026.02 | 15.2 | 43 | 56.5 | |
| Visual Model Checking + CLIPZero-Shot=true2026.02 | 4.3 | 26.1 | 56.5 |