VLEP (Video-and-Language Event Prediction)

VLEP 包含来自 10,234 个不同的电视节目和 YouTube Lifestyle Vlog 视频剪辑的 28,726 个未来事件预测示例（及其基本原理）。每个示例（参见图 1）由一个前提事件（一个带有对话的短视频剪辑）、一个前提摘要（前提事件的文本摘要）和两个由人们编写的潜在自然语言未来事件（以及基本原理）组成。这些剪辑的平均长度为 6.1 秒，取自各种事件丰富的来源，例如电视节目和 YouTube Lifestyle Vlog 视频。

VLEP contains 28,726 future event prediction examples (along with their rationales) sourced from 10,234 distinct television programs and YouTube Lifestyle Vlog video clips. Each example (see Figure 1) consists of a premise event (a short video clip with dialogue), a premise summary (the textual summary of the premise event), and two human-written potential natural language future events accompanied by their rationales. These clips have an average duration of 6.1 seconds, and are collected from a variety of event-rich sources such as television programs and YouTube Lifestyle Vlog videos.

搜集汇总

背景与挑战

背景概述

VLEP是一个用于视频和语言事件预测的数据集，包含28,726个未来事件预测示例，这些示例基于来自电视节目和YouTube生活方式视频博客的10,234个短视频剪辑。每个示例由前提事件、前提摘要以及两个潜在未来事件及其基本原理组成，视频剪辑平均时长为6.1秒。