README.md

Question-Relevance-in-VQA

Project for 10-805: Machine Learning for large datasets.

Final report.

Data:

Visual vs Non Visual Data

    Visual Questions:
        VQA 2.0: 381,055
    Non visual questions:
        Squad: 97,868
        VQA 2.0: 9,949

True vs False Premise Data

    QRPE Dataset:
        True Premise/Relevant: (34324+17738) = 52062
        False Premise/Non Relevant: (35392+18372) = 53764
        Total: 105826
    
    Extended Dataset:
        True Premise/Relevant: (443757+214354) = 658111
        False Premise/Non Relevant: (1319368+693558) = 2012926
        Total: 2671037