We introduce VQAonline, the first VQA dataset in which all contents originate from an authentic use case.
VQAonline includes 64K visual questions sourced from an online question answering community (i.e., StackExchange).
It differs from prior datasets; examples include that it contains: