Visual Commonsense Reasoning on VCR e-ViL (test)
26.4Meteor ScoreNLX-GPT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NLX-GPT2022.12 | 26.4 | 80.3 | — | — | |
| FME2022.12 | 22.7 | 79.4 | 58.5 | 48.9 | |
| e-UG2022.12 | 22.5 | 79 | 65.1 | 69.8 | |
| PJ-X2022.12 | 20.5 | 78.4 | 52.7 | 39 | |
| OFA-Xmodel_variant=large (470M parameters), training=single-task fine-tuned2022.12 | 12.2 | 79.8 | 68.9 | 71.2 | |
| OFA-XMTmodel_variant=large (470M parameters), training=multi-task pre-training (e-ViL-combined)2022.12 | 11.3 | 79.3 | 77.3 | 62 | |
| Ground-truth2022.12 | — | — | 87.6 | — |