Negation Understanding on Neg-COCO MCQ
48.6AccuracyDCSM
Evaluation results
| Method | Links | |
|---|---|---|
| DCSMTraining Source=coco, Base Model=CLIP ViTB/162025.03 | 48.6 | |
| CLIPBackbone=ViT B/162025.03 | 41.5 | |
| CLIPBackbone=ViT B/322025.03 | 39.2 | |
| DCSMTraining Source=synth, Base Model=CLIP ViTB/162025.03 | 38.8 | |
| CLIPGlassesTraining Set=Neg-COCO-R, Training Size (Images/Captions)=5K/22K2026.02 | 35.9 | |
| CLIPGlassesTraining Set=CC-Neg, Training Size (Images/Captions)=188K/376K2026.02 | 34.51 | |
| NegCLIP2025.03 | 31.4 | |
| CoN-CLIPTraining Set=Neg-COCO-R, Training Size (Images/Captions)=5K/22K2026.02 | 30.61 | |
| SigLIP2025.03 | 26.3 | |
| CoN-CLIPTraining Set=CC-Neg, Training Size (Images/Captions)=188K/376K2026.02 | 25.7 | |
| Random Chance2025.03 | 25 | |
| CoCa2025.03 | 21.6 | |
| BLIP2025.03 | 20.5 | |
| NegCLIPTraining Set=COCO-caption, Training Size (Images/Captions)=82K/31K2026.02 | 10.2 |