MovieFIB (Movie Fill-in-the-Blank)

问答已成为一项流行的任务，具有许多实际应用（例如对话系统）。它非常容易解释和定量评估，并且通过简单的设置，生成适用于深度学习的非常大的数据集相对容易。大规模数据集的视觉 QA（图像和自然语言问题）的进步推动了该领域的快速发展，我们希望通过这个挑战数据集将这一进展扩展到视频。

Question Answering (QA) has emerged as a popular task with numerous practical applications, such as dialogue systems. It is straightforward to interpret and quantitatively evaluate, and generating extremely large datasets suitable for deep learning is relatively easy with simple setup configurations. Advances in large-scale Visual Question Answering (VQA) datasets—where tasks pair images with natural language questions—have fueled rapid progress in this field, and we intend to extend this progress to the video domain via this challenging benchmark dataset.

搜集汇总

背景与挑战

背景概述

MovieFIB是一个专注于视频问答的数据集，采用填空形式来促进对视频内容的理解。它于2017年发布，旨在将视觉问答领域的进展扩展到视频数据中。