Image-to-Text Retrieval on MS-COCO (val)
0.601R@1DINOv2-ARL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DINOv2-ARLTraining Samples=20M2024.09 | 0.601 | — | — | |
| LAION-CLIP ViT-LBackbone=ViT-L, Training Samples=400M2024.09 | 0.597 | — | — | |
| DINOv2-MpNetBackbone=MpNet, Training Samples=20M2024.09 | 0.58 | — | — | |
| OpenAI-CLIP ViT-LBackbone=ViT-L, Training Samples=400M2024.09 | 0.563 | — | — | |
| LiT L16LBackbone=L16L, Training Samples=112M2024.09 | 0.485 | — | — | |
| LilTLwA-baseBackbone=base, Training Samples=0.5M2024.09 | 0.47 | — | — | |
| LilTDA-baseBackbone=base, Training Samples=0.5M2024.09 | 0.414 | — | — | |
| CLIPw/ boxes=false2025.10 | — | 0.7032 | 0.8009 | |
| CLIPw/ boxes=true2025.10 | — | 0.7035 | 0.8019 | |
| HyCoCLIPw/ boxes=true2025.10 | — | 0.6951 | 0.7973 | |
| MERUw/ boxes=false2025.10 | — | 0.6957 | 0.7933 | |
| MERUw/ boxes=true2025.10 | — | 0.697 | 0.7969 | |
| PHyCLIPw/ boxes=true2025.10 | — | 0.7094 | 0.8086 |