A GuessWhat?! Game for Goal-Oriented Visual Dialog: A Survey

Visual Dialog requires an AI-agent to hold a meaningful natural conversation about visual content. For a given image, dialog history, and a current question about the image, the AI-agent is required to understand the question, infer the context from the history and ground this information from the given image to be able to answer the current question correctly. One of the Visual Dialog tasks is a guessing game between two AI-agents for visual object discovery. This work represents a survey over different approaches done for the guessing game task. In this survey, we are showing the methodologies and architectures used by the relevant work as well as the results achieved by them. We present the most recent five approaches for this game from 2017 to 2020. GuessWhat?! is the dataset, introduced mainly for addressing the guessing game for visual dialog task, used in all works we are mentioning here.