Attention-based Visual Question Generation

Visual questions are the questions asked with reference to visual content viz. images and videos. Generating visual questions is a challenging task as, it involves asking semantically valid and syntactically correct questions. Most of the methods for visual question generation use the rule-based or template-based approaches. For the same reason they fail to generate novel questions. In this paper, we propose an attention-based mechanism for generating visual questions. The paper compares the results of a simpler encoder-decoder model with the attention model and show the effectiveness of attention for the said task. It highlights the explainable nature of attention mechanism for generating visual questions.