Text-to-Image Retrieval on Wikipedia
47.8R@1GRAPE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRAPEBackbone=ViT-L/14, Method=+GRAPE2025.09 | 47.8 | 83.5 | |
| GRAPEBackbone=ViT-B/16, Method=+GRAPE2025.09 | 43.7 | 82.8 | |
| CLIP + LLMBackbone=ViT-L/14, Method=+LLM2025.09 | 42 | 77.6 | |
| CLIPBackbone=ViT-L/14, Method=Vanilla CLIP2025.09 | 40.4 | 80.2 | |
| GRAPEBackbone=ViT-B/32, Method=+GRAPE2025.09 | 36.1 | 78.4 | |
| CLIP + LLMBackbone=ViT-B/16, Method=+LLM2025.09 | 34.3 | 69.3 | |
| CLIPBackbone=ViT-B/16, Method=Vanilla CLIP2025.09 | 33.5 | 71.9 | |
| CLIP + LLMBackbone=ViT-B/32, Method=+LLM2025.09 | 30.4 | 67.1 | |
| CLIPBackbone=ViT-B/32, Method=Vanilla CLIP2025.09 | 28.4 | 65.8 |