Text-to-Image Retrieval on UCM (val)
66.6Recall@5CLIP (oracle)
Evaluation Results
| Method | Links | |
|---|---|---|
| CLIP (oracle)Data=L + U2026.04 | 66.6 | |
| ToMA-domainData=L=U2026.04 | 63 | |
| ToMA-domainData=L≠U2026.04 | 61.3 | |
| ToMAData=L=U2026.04 | 60.8 | |
| ToMAData=L≠U2026.04 | 60.5 | |
| SemiCLIPData=L=U2026.04 | 59.4 | |
| SemiCLIPData=L≠U2026.04 | 59.4 | |
| S-CLIPData=L=U2026.04 | 57.3 | |
| CLIP (fine-tuned)Data=L2026.04 | 56.3 | |
| S-CLIPData=L≠U2026.04 | 55.7 | |
| Soft-PLData=L=U2026.04 | 55.6 | |
| Hard-PLData=L=U2026.04 | 54.6 | |
| Soft-PLData=L≠U2026.04 | 54.4 | |
| Hard-PLData=L≠U2026.04 | 54.3 | |
| S-CLIPData setup=L=U2023.05 | 43.5 | |
| Soft-PLData setup=L=U2023.05 | 42.9 | |
| S-CLIPData setup=L!=U2023.05 | 42.5 | |
| Hard-PLData setup=L!=U2023.05 | 42.1 | |
| CLIP (fine-tune)Data setup=L2023.05 | 40.9 | |
| Soft-PLData setup=L!=U2023.05 | 40.8 | |
| Hard-PLData setup=L=U2023.05 | 40.5 | |
| CLIP (original)Data=-2026.04 | 39.5 | |
| CLIP (original)Setup=Original2023.05 | 24.8 |