Image-text alignment on Winoground (test)
89.5Text ScoreHumans
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Humans2023.05 | 89.5 | 88.5 | 85.5 | |
| VQ2mode=zero-shot2023.05 | 47 | 42.2 | 30.5 | |
| PaLIFine-tuning dataset=SNLI-VE + Synthetic Data2023.05 | 46.5 | 38 | 28.75 | |
| PaLIFine-tuning dataset=SNLI-VE2023.05 | 45 | 41.5 | 28.7 | |
| BLIP2Fine-tuning dataset=COCO2023.05 | 44 | 26 | 23.5 | |
| IAISlarge2023.05 | 42.5 | 19.75 | 16 | |
| VinVL2023.05 | 37.75 | 17.75 | 14.5 | |
| COCA VIT-L/14Fine-tuning dataset=COCO, Backbone=VIT-L/142023.05 | 28.25 | 11.5 | 8.25 | |
| OFA LargeFine-tuning dataset=SNLI-VE2023.05 | 27.7 | 14.3 | 9 | |
| CLIP RN50x64Backbone=RN50x642023.05 | 26.5 | 13.75 | 10.25 | |
| Random Chance2023.05 | 25 | 25 | 16.67 | |
| TIFA2023.05 | 19 | 12.5 | 11.3 |