Visually Grounded Commonsense Knowledge Acquisition