Relation Prediction on VG8K-LT v1.4 (test)
37.3Accuracy (many)RelTransformer
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RelTransformerLearning Method=DCPL2021.04 | 37.3 | 27.6 | 16.5 | 19.2 | |
| RelTransformerLearning Method=WCE2021.04 | 36.6 | 27.4 | 16.3 | 19 | |
| LSVRULearning Method=WCE2021.04 | 35.5 | 24.7 | 15.2 | 17.2 | |
| LSVRULearning Method=DCPL2021.04 | 34.3 | 15.1 | 12.9 | 14.4 | |
| RelTransformerLearning Method=Focal Loss2021.04 | 30.5 | 22.8 | 14.8 | 16.8 | |
| VilHubArchitecture=LSVRU2021.04 | 27.5 | 17.4 | 14.6 | 15.7 | |
| RelTransformerLearning Method=CE2021.04 | 26.8 | 18.6 | 15 | 16.1 | |
| VilHub + RelMixArchitecture=LSVRU2021.04 | 24.5 | 16.5 | 14.4 | 15.4 | |
| LSVRULearning Method=Focal Loss2021.04 | 24.5 | 16.2 | 13.7 | 14.7 | |
| EQLArchitecture=LSVRU2021.04 | 22.6 | 15.6 | 12.6 | 13.6 | |
| OLTRArchitecture=LSVRU2021.04 | 22.5 | 15.6 | 12.6 | 13.6 | |
| LSVRULearning Method=CE2021.04 | 22.2 | 15.5 | 12.6 | 13.5 | |
| CounterfactualArchitecture=LSVRU, reproduction=true2021.04 | 12.1 | 25.6 | 14.9 | 17.1 |