Image-to-Text Retrieval on ShareGPT4V 5,000 samples (test)
88.62R@1CLIP+GOAL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CLIP+GOALBackbone=ViT-L/142025.03 | 88.62 | 97.88 | 99.76 | 99.92 | |
| CLIP+GOALBackbone=ViT-B/162025.03 | 85.44 | 97.12 | 99.62 | 99.84 | |
| CLIP+LongCLIPBackbone=ViT-B/162025.03 | 75.56 | 93.36 | 98.78 | 99.62 | |
| CLIP+LongCLIPBackbone=ViT-L/142025.03 | 73.84 | 91.44 | 98.5 | 99.48 | |
| CLIPBackbone=ViT-L/142025.03 | 62.7 | 81.78 | 93.78 | 96.64 | |
| CLIPBackbone=ViT-B/162025.03 | 62.24 | 82.32 | 95 | 97.74 |