Compositional Reasoning on SUGARCREPE (test)
86.3AccuracyCLIP+TF_Local
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CLIP+TF_LocalTraining data=TROHN-Img2026.04 | 86.3 | 77.1 | |
| FSC-CLIPTraining data=COCO†2026.04 | 85.1 | 77.6 | |
| NegCLIPTraining data=COCO†2026.04 | 83.6 | 76.6 | |
| TripletCLIPTraining data=CC3M/12M2026.04 | 82.6 | 71.9 | |
| X-VLMTraining data=COCO+Others2026.04 | 81.9 | 67.4 | |
| CLIP+TF_LocalTraining data=COCO2026.04 | 80.9 | 76.3 | |
| FineCLIPTraining data=COCO†2026.04 | 80.6 | 68.2 | |
| CLIPTraining data=Pretrained2026.04 | 73 | 63.3 |