Fine-grained Image-Text Matching on Winoground
25.8Group AgreementX2-VLMlarge
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| X2-VLMlargeMode=Zero-shot, Backbone=Large2022.11 | 25.8 | 52.5 | 32.5 | |
| X2-VLMlarge4MMode=Zero-shot, Backbone=Large, Pre-training Data=4M2022.11 | 25.5 | 49.5 | 31 | |
| X2-VLMbaseMode=Zero-shot, Backbone=Base2022.11 | 24.5 | 47.3 | 29.8 | |
| X2-VLMbase4MMode=Zero-shot, Backbone=Base, Pre-training Data=4M2022.11 | 22.5 | 46.3 | 25.3 | |
| BLIP-2Mode=Zero-shot2022.11 | 18.2 | 43 | 22 | |
| RandomMode=Zero-shot2022.11 | 12.5 | 25 | 25 | |
| PEVLMode=Zero-shot2022.11 | 12.2 | 33.2 | 15.7 | |
| BLIPMode=Zero-shot2022.11 | 11.7 | 35.5 | 15 | |
| ALBEF4MMode=Zero-shot, Pre-training Data=4M2022.11 | 11 | 29.2 | 15.5 | |
| UNITERlargeMode=Zero-shot, Backbone=Large2022.11 | 10.5 | 38 | 14 | |
| CLIPMode=Zero-shot2022.11 | 8 | 30.7 | 10.5 |