Relation Prediction on GQA-LT (test)
63.6Accuracy (many)VilHub
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VilHubArchitecture=LSVRU2021.04 | 63.6 | 17.6 | 7.2 | 11.7 | |
| RelTransformerLearning Method=WCE2021.04 | 63.6 | 59.1 | 43.1 | 46.5 | |
| OLTRArchitecture=LSVRU2021.04 | 63.5 | 15 | 8.2 | 12.1 | |
| VilHub + RelMixArchitecture=LSVRU2021.04 | 63.4 | 14.9 | 8 | 11.9 | |
| RelTransformerLearning Method=CE2021.04 | 63.4 | 16.6 | 7 | 11.2 | |
| LSVRULearning Method=CE2021.04 | 62.6 | 15.5 | 6.8 | 11 | |
| EQLArchitecture=LSVRU2021.04 | 62.3 | 15.8 | 6.6 | 10.8 | |
| RelTransformerLearning Method=Focal Loss2021.04 | 61.9 | 16.8 | 8.3 | 12.2 | |
| LSVRULearning Method=DCPL2021.04 | 61.4 | 23.6 | 7.6 | 12.7 | |
| LSVRULearning Method=Focal Loss2021.04 | 60.4 | 15.7 | 7.7 | 11.6 | |
| RelTransformerLearning Method=DCPL2021.04 | 58.4 | 38.6 | 13.2 | 19.3 | |
| LSVRULearning Method=WCE2021.04 | 53.4 | 35.1 | 15.7 | 20.5 | |
| CounterfactualArchitecture=LSVRU, reproduction=true2021.04 | 38.6 | 38 | 9.4 | 15.2 |