Image-to-Text Retrieval on KAIST (test)
7.8Recall@1T-CLIP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| T-CLIPGranularity=Dual2026.05 | 7.8 | 25.2 | 37.4 | 57.1 | 72.9 | 86.2 | |
| T-CLIPGranularity=Global2026.05 | 7.1 | 23.3 | 34.9 | 54.1 | 69.5 | 83.7 | |
| Global LoRAGranularity=Global2026.05 | 7 | 22.4 | 33.2 | 52 | 68.5 | 82.5 | |
| DeCLIPGranularity=Global2026.05 | 3.8 | 12.9 | 20.6 | 34.9 | 49.7 | 64.4 | |
| F-G LoRAGranularity=Fine-grained2026.05 | 2.1 | 7.3 | 12.5 | 24.8 | 38.1 | 57.1 | |
| CLIP-AdapterGranularity=Global2026.05 | 1.8 | 7.2 | 12.3 | 22.9 | 35.7 | 52.1 | |
| T-CLIPGranularity=Fine-grained2026.05 | 1.6 | 6.4 | 10.1 | 19.4 | 31.1 | 47.3 | |
| DeCLIPGranularity=Fine-grained2026.05 | 1.3 | 4.8 | 8.8 | 16.9 | 27.3 | 43.3 | |
| CLIP-AdapterGranularity=Fine-grained2026.05 | 0.9 | 4.1 | 8.1 | 16.7 | 28.3 | 43.6 | |
| CLIPGranularity=Global2026.05 | 0.3 | 1.6 | 3.4 | 6.8 | 11 | 17.9 | |
| CLIPGranularity=Fine-grained2026.05 | 0.1 | 0.4 | 0.8 | 2.9 | 5.7 | 9.6 |