GOAL
收藏资源简介:
GOAL数据集由清华大学计算机科学与技术系创建,包含超过8.9k个足球视频片段,旨在支持实时足球评论生成。数据集中的每个视频都与来自专业体育平台和Wikidata的丰富知识相关联。创建过程中,通过一系列人工标注任务,如校对、实体识别和文本分类,完成了知识、视频和文本的精细对齐。该数据集的应用领域主要集中在自动体育叙述,旨在解决如何基于背景知识生成生动、细粒度的视频描述问题。
The GOAL dataset was created by the Department of Computer Science and Technology, Tsinghua University. It contains over 8.9 thousand football video clips, and is designed to support real-time football commentary generation. Each video in the dataset is associated with rich knowledge sourced from professional sports platforms and Wikidata. During its development, fine-grained alignment among knowledge, videos and texts was accomplished via a series of manual annotation tasks, including proofreading, entity recognition and text classification. The core application areas of this dataset focus on automated sports narration, aiming to address the challenge of generating vivid, fine-grained video descriptions based on background knowledge.

- 1GOAL: A Challenging Knowledge-grounded Video Captioning Benchmark for Real-time Soccer Commentary Generation清华大学计算机科学与技术系 · 2023年



