Compositionality Evaluation on VLC
72.6Average ScoreMosaiCLIP
Evaluation Results
| Method | Links | |
|---|---|---|
| MosaiCLIPArch.=RN-50, Pre-training data=CC-12M2023.05 | 72.6 | |
| MosaiCLIPArch.=Swin-T, Pre-training data=CC-12M2023.05 | 72.4 | |
| MosaiCLIPArch.=RN-50, Pre-training data=YFCC-15M2023.05 | 72.3 | |
| MosaiCLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.05 | 71.8 | |
| NegCLIPArch.=Swin-T, Pre-training data=CC-12M2023.05 | 70 | |
| NegCLIPArch.=RN-50, Pre-training data=CC-12M2023.05 | 69.5 | |
| CLIPArch.=RN-50, Pre-training data=CC-12M2023.05 | 66.2 | |
| CLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.05 | 66.2 | |
| NegCLIPArch.=RN-50, Pre-training data=YFCC-15M2023.05 | 66.1 | |
| NegCLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.05 | 66 | |
| CLIPArch.=Swin-T, Pre-training data=CC-12M2023.05 | 65.6 | |
| CLIPArch.=RN-50, Pre-training data=YFCC-15M2023.05 | 64.8 | |
| RandomPre-training data=CC-12M2023.05 | 50 | |
| RandomPre-training data=YFCC-15M2023.05 | 50 |