Overcoming Language Priors for Visual Question Answering Based on Knowledge Distillation

Daowan Peng, Wei Wei

Previous studies have pointed out that visual question answering (VQA) models are prone to relying on language priors for answer predictions. In this context, predictions often dep…