Visual Question Answering on VizWiz 2022 (test-std)
60.15AccuracyEnsemble
Evaluation Results
| Method | Links | |
|---|---|---|
| EnsembleComponents=RN50x64 + ViT-L/14@336px2022.06 | 60.15 | |
| RN50x64Backbone=RN50x64 (CLIP), Resolution=448x448, Frozen Backbone=true2022.06 | 59.4 | |
| ViT-L/14@336pxBackbone=ViT-L/14 (CLIP), Resolution=336x336, Frozen Backbone=true2022.06 | 59.01 |