Instance-level Text-to-Image Retrieval on COCO + LAION
0.485R@1ViT-L/14FT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ViT-L/14FTGFLOPs=175.332025.12 | 0.485 | 0.735 | 0.818 | |
| ICAR-20/24GFLOPs=164.412025.12 | 0.474 | 0.727 | 0.811 | |
| ICAR-16/24GFLOPs=151.052025.12 | 0.471 | 0.722 | 0.812 | |
| ICAR-12/24GFLOPs=137.682025.12 | 0.451 | 0.702 | 0.793 | |
| ICAR-8/24GFLOPs=124.312025.12 | 0.429 | 0.683 | 0.774 |