Visual Entailment on SNLI-VE (test-p)
80.2AccuracyCLIP-ViLp
Evaluation Results
| Method | Links | |
|---|---|---|
| CLIP-ViLpVisual Encoder=CLIP-Res50x4, V&L Pretrain Data=9.2M, V&L Pretrain Epoch=202021.07 | 80.2 | |
| CLIP-ViLpVisual Encoder=CLIP-Res50, V&L Pretrain Data=9.2M, V&L Pretrain Epoch=202021.07 | 78.97 | |
| UNITERVisual Encoder=BUTD-Res101, V&L Pretrain Data=6.5M2021.07 | 78.28 | |
| VL-HyperPELTNumber of Samples=20002022.03 | 65.67 | |
| VL-AdapterNumber of Samples=20002022.03 | 64.22 | |
| VL-HyperPELTNumber of Samples=10002022.03 | 63.01 | |
| VL-AdapterNumber of Samples=10002022.03 | 62.83 | |
| CLIP-T5Number of Samples=20002022.03 | 62.29 | |
| VL-HyperPELTNumber of Samples=5002022.03 | 61.97 | |
| VL-AdapterNumber of Samples=5002022.03 | 60.58 | |
| CLIP-T5Number of Samples=10002022.03 | 60.49 | |
| CLIP-T5Number of Samples=5002022.03 | 59.01 | |
| VL-HyperPELTNumber of Samples=1002022.03 | 57.69 | |
| VL-AdapterNumber of Samples=1002022.03 | 55.05 | |
| VL-HyperPELTNumber of Samples=322022.03 | 53.4 | |
| VL-AdapterNumber of Samples=322022.03 | 51.69 | |
| CLIP-T5Number of Samples=1002022.03 | 49.97 | |
| VL-HyperPELTNumber of Samples=162022.03 | 47.86 | |
| VL-AdapterNumber of Samples=162022.03 | 46.67 | |
| VL-HyperPELTNumber of Samples=42022.03 | 39.94 | |
| CLIP-T5Number of Samples=322022.03 | 39.89 | |
| VL-AdapterNumber of Samples=42022.03 | 38.75 | |
| CLIP-T5Number of Samples=162022.03 | 38.55 | |
| CLIP-T5Number of Samples=42022.03 | 34.77 |