Image-to-Text Retrieval on COCO-CN (test)
79.3R@1R2D2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R2D2Backbone=ViT-L2022.05 | 79.3 | 97.1 | 98.7 | 91.6 | |
| R2D2Backbone=ViT-B2022.05 | 78.1 | 96.2 | 98.6 | 90.3 | |
| WukongBackbone=ViT-L2022.05 | 73.3 | 94 | 98 | 88.6 | |
| FILIPBackbone=ViT-L2022.05 | 69.1 | 91.3 | 96.9 | 86.5 | |
| CLIPBackbone=ViT-B2022.05 | 68.7 | 93.6 | 97.5 | 86.6 | |
| CLIPBackbone=ViT-L2022.05 | 68.3 | 93 | 97.3 | 86.2 | |
| WukongBackbone=ViT-B2022.05 | 65.8 | 90.3 | 96.6 | 84.6 | |
| FILIPBackbone=ViT-B2022.05 | 52.7 | 81.3 | 88.3 | 76.6 |