Visual Commonsense Reasoning on VCR (dev)
70.8Q -> A AccuracyVisualBERT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VisualBERTEarly Fusion=true, COCO Pre-training=true2019.08 | 70.8 | 73.2 | 52.2 | |
| VisualBERT w/o Early FusionEarly Fusion=false2019.08 | 70.1 | 71.9 | 50.6 | |
| VisualBERT w/o COCO Pre-trainingCOCO Pre-training=false2019.08 | 67.9 | 69.5 | 47.9 | |
| R2C2019.08 | 63.8 | 67.2 | 43.1 |