BLOCK: Bilinear Superdiagonal Fusion for Visual Question Answering and Visual Relationship Detection

Benchmark Model Rank Results
visual-relationship-detection-on-vrd-phrase-detectionBLOCK#1R@100: 28.96R@50: 26.32
visual-relationship-detection-on-vrd-predicate-detectionBLOCK#2R@100: 92.58R@50: 86.58
visual-relationship-detection-on-vrd-relationship-detectionBLOCK#2R@100: 20.96R@50: 19.06
visual-question-answering-on-vqa-v2-test-stdBLOCK#33overall: 67.9
visual-question-answering-on-vqa-v2-test-devBLOCK#39Accuracy: 67.58