Soccer captioning dataset
收藏资源简介:
Soccer captioning dataset是由比利时ISIA实验室创建的一个包含22,000个足球视频-标题对的数据集,用于深度学习模型的训练和评估。该数据集从SoccerNet视频中提取,标题则从flashscore.com网站爬取并格式化处理。数据集涵盖了多种足球动作,如射门、角球、换人等,每种动作的标题数量在数据集中有所体现。创建过程中,视频通过提取图像、光流和图像修复三种视觉特征进行处理,以降低数据维度并提供更多视频线索。该数据集主要应用于足球视频自动标题生成领域,旨在通过深度学习技术模仿人类评论员,提供准确且多样化的视频描述。
The Soccer Captioning Dataset is a collection of 22,000 football video-caption pairs developed by the ISIA Laboratory of Belgium, intended for training and evaluating deep learning models. This dataset is extracted from SoccerNet videos, while its captions are crawled from the flashscore.com website and subsequently formatted and processed. The dataset covers a wide range of football actions including shots, corner kicks, substitutions and more, with the count of captions for each action recorded within the dataset. During the dataset creation process, videos are processed by extracting three types of visual features: images, optical flow, and image restoration, to reduce data dimensionality and provide additional video cues. This dataset is primarily utilized in the domain of automatic football video caption generation, with the goal of mimicking human commentators via deep learning technologies to generate accurate and diverse video descriptions.

- 1Deep soccer captioning with transformer: dataset, semantics-related losses, and multi-level evaluation比利时ISIA实验室 · 2022年



