Image-to-Text Generation on MS-COCO 5% 2014 (test val)
25.5CLIP Similarity (CLIP)GIT-large
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GIT-largeAlpha level (α)=0.1, Model scale=large, Image size=224 × 224, Maximum text length=502026.05 | 25.5 | -0.0043 | -0.2593 | 86.49 | 64.59 | -0.219 | |
| GIT-baseAlpha level (α)=0.1, Model scale=base, Image size=224 × 224, Maximum text length=502026.05 | 25.11 | -0.0021 | -0.2532 | 86.2 | 64.74 | -0.2146 | |
| BLIP-largeAlpha level (α)=0.1, Model scale=large, Image size=224 × 224, Maximum text length=502026.05 | 24.53 | -0.0074 | -0.2527 | 81.06 | 56.31 | -0.2475 | |
| BLIP-baseAlpha level (α)=0.1, Model scale=base, Image size=224 × 224, Maximum text length=502026.05 | 23.3 | 0.007 | -0.226 | 83.49 | 63.35 | -0.2014 |