Image-Text Compositionality Evaluation on SugarCrepe ++ (test)
79.7Replace ITTC2LIP
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| C2LIPMethod Category=Models fine-tuned by us2026.03 | 79.7 | — | — | — | — | — | — | — | — | — | — | 75.3 | 55.2 | 44.2 | |
| CLICMethod Category=Composition-aware models2026.03 | 76.6 | — | — | — | — | — | — | — | — | — | — | 58.1 | 59 | 27 | |
| FG-CLIPMethod Category=Fine-grained models2026.03 | 75.8 | — | — | — | — | — | — | — | — | — | — | 67.5 | 51.5 | 38.2 | |
| SigLIPBackbone=ViT-B/162026.03 | 73.8 | — | — | — | — | — | — | — | — | — | — | 62.8 | 48 | 34.7 | |
| SigLIPBackbone=ViT-B/16, Fine-tuning Dataset=CC3M, Method Category=Models fine-tuned by us2026.03 | 73.5 | — | — | — | — | — | — | — | — | — | — | 67.9 | 49.8 | 36.6 | |
| FineCLIPMethod Category=Fine-grained models2026.03 | 72.8 | — | — | — | — | — | — | — | — | — | — | 68.7 | 43.9 | 26.8 | |
| CLIP (OpenAI)Backbone=ViT-B/162026.03 | 70.1 | — | — | — | — | — | — | — | — | — | — | 60.2 | 43.8 | 23.9 | |
| TripletCLIPMethod Category=Composition-aware models2026.03 | 69.9 | — | — | — | — | — | — | — | — | — | — | 69.4 | 41.4 | 28.5 | |
| CLIP (OpenAI)Backbone=ViT-B/322026.03 | 69.5 | — | — | — | — | — | — | — | — | — | — | 60.5 | 45.7 | 27.4 | |
| NegCLIPMethod Category=Composition-aware models2026.03 | 69.1 | — | — | — | — | — | — | — | — | — | — | 70.9 | 53.4 | 39.1 | |
| CoN-CLIPMethod Category=Composition-aware models2026.03 | 68.1 | — | — | — | — | — | — | — | — | — | — | 70.1 | 45.2 | 28 | |
| LLIPMethod Category=Fine-grained models2026.03 | 67.3 | — | — | — | — | — | — | — | — | — | — | 56.8 | 40.8 | 27.3 | |
| FLAIR-3mMethod Category=Fine-grained models2026.03 | 66.3 | — | — | — | — | — | — | — | — | — | — | 57.5 | 49.5 | 35.2 | |
| SLVC-RMethod Category=Composition-aware models2026.03 | 64 | — | — | — | — | — | — | — | — | — | — | 68.2 | 51.3 | 28.4 | |
| SLVC-RLMethod Category=Composition-aware models2026.03 | 61.9 | — | — | — | — | — | — | — | — | — | — | 70 | 46 | 26.3 | |
| DreamLIP-3mMethod Category=Fine-grained models2026.03 | 61.2 | — | — | — | — | — | — | — | — | — | — | 51.5 | 44.4 | 30.1 | |
| CE-CLIPMethod Category=Composition-aware models2026.03 | 56.5 | — | — | — | — | — | — | — | — | — | — | 67 | 34.3 | 32.5 | |
| DAC-LLMMethod Category=Composition-aware models2026.03 | 53.7 | — | — | — | — | — | — | — | — | — | — | 59.6 | 32.2 | 18.1 | |
| DAC-SAMMethod Category=Composition-aware models2026.03 | 52.4 | — | — | — | — | — | — | — | — | — | — | 60.2 | 30.6 | 18.4 | |
| IL-CLIPMethod Category=Codebook-based models2026.03 | 34.4 | — | — | — | — | — | — | — | — | — | — | 31.8 | 36.9 | 14.7 | |
| Codebook-CLIPMethod Category=Codebook-based models2026.03 | 34.3 | — | — | — | — | — | — | — | — | — | — | 13.8 | 28.3 | 10.1 | |
| BLIPBackbone=ViT-L, Params (B)=0.232024.12 | — | 46.8 | 29.8 | 60.1 | 52.5 | 92.6 | 89.1 | 71.7 | 75 | 56.8 | 57.7 | — | — | — | |
| BLIP-SGVLParams (B)=0.152024.12 | — | 13.2 | — | — | 38.8 | 53.8 | 34.4 | — | — | 30.7 | — | — | — | — | |
| BLIP2Backbone=ViT-L, Params (B)=1.172024.12 | — | 37.9 | 39.5 | 51.9 | 55.4 | 94.8 | 96.9 | 73.2 | 86.5 | 65.1 | 69.6 | — | — | — | |
| CLIPBackbone=ViT-L, Params (B)=0.432024.12 | — | 46 | 14.5 | 44.5 | 28.7 | 92 | 81.3 | 68.8 | 56.3 | 53.4 | 39.1 | — | — | — | |
| CLIP-SVLCParams (B)=0.152024.12 | — | 43 | 18.9 | 48.4 | 34.6 | 80.9 | 91.6 | 57 | 66.9 | 47.3 | 51.3 | — | — | — | |
| E5-VBackbone=LLAVA-Next-8B, Params (B)=8.362024.12 | — | 50.8 | 48.4 | 49.7 | 56.9 | 93.1 | 97.6 | 76.1 | 87.1 | 74.7 | 84.4 | — | — | — | |
| E5-VBackbone=LLAVA-1.5-7B, Params (B)=7.062024.12 | — | 39.5 | 42.3 | 40.7 | 48.5 | 89.7 | 94.6 | 71.7 | 86.4 | 72 | 81.5 | — | — | — | |
| EVA-02-CLIPBackbone=ViT-E/14+, Params (B)=5.042024.12 | — | 48.4 | 28.2 | 56.3 | 49.4 | 94.5 | 88.9 | 76.3 | 70.6 | 59.4 | 49.4 | — | — | — | |
| EVA-CLIPParams (B)=8.222024.12 | — | 43.6 | 25.4 | 55.2 | 46.9 | 93.7 | 85.8 | 73.4 | 67.9 | 59.7 | 49.2 | — | — | — | |
| EVA-CLIPParams (B)=18.32024.12 | — | 45.2 | 25.4 | 55.5 | 47.6 | 94.1 | 85.1 | 77 | 69.8 | 60.4 | 47.8 | — | — | — | |
| Human2024.12 | — | 100 | 96.7 | 96.7 | 93.3 | 100 | 97 | 100 | 98.3 | 100 | 96.7 | — | — | — | |
| LLaVA-1.5-7BParams (B)=7.062024.12 | — | 23.8 | 30.7 | 28 | 29.5 | 58.1 | 63 | 46.8 | 58.1 | 52.3 | 63.4 | — | — | — | |
| NegCLIPParams (B)=0.152024.12 | — | 55.3 | 34.7 | 58 | 56.5 | 89.5 | 94.5 | 69.4 | 76.3 | 52.3 | 51.6 | — | — | — | |
| OpenCLIPBackbone=ViT-G/14, Params (B)=1.372024.12 | — | 40.7 | 27.4 | 54.2 | 49.6 | 93.1 | 89.4 | 72.5 | 73.1 | 57.6 | 51.4 | — | — | — | |
| OpenCLIPBackbone=ViT-BigG/14, Params (B)=2.542024.12 | — | 48.8 | 28.2 | 57.7 | 52.4 | 94.2 | 90.5 | 76.4 | 72.6 | 59.4 | 53.6 | — | — | — | |
| VladVABackbone=LLAVA-1.5-7B, Params (B)=7.062024.12 | — | 56.1 | 36.7 | 63 | 62.5 | 95 | 93 | 78.2 | 82.3 | 71.1 | 66.3 | — | — | — | |
| VLM2VecBackbone=Mistral-7B, Params (B)=7.302024.12 | — | 40.7 | 39.9 | 48.1 | 50 | 94.6 | 96.9 | 77 | 85.6 | 67.9 | 70.7 | — | — | — |