MCG-NJU/SportsGrounding

SportsGrounding数据集基于MultiSports数据集的篮球子集，专注于篮球相关场景。它包含了许多复杂的人与人、人与物体之间的交互场景，实例数量较多。数据集经过修改，移除了六个动作重复过多且无法通过自然语言区分的视频，最终包含520个视频，其中训练集374个，验证集146个。与其他STVG数据集不同，SportsGrounding中的每个视频包含多个描述不同目标人物的字幕。数据集的关键统计数据包括4243个视频-文本对、平均视频时长19.70秒、平均管时长1.49秒和平均描述长度16.89个词。SportsGrounding的独特之处在于一些实例的持续时间非常短，视频中出现的人更多，人与人之间的交互更复杂，许多描述需要一定的推理和场景信息建模。

The SportsGrounding dataset is based on the basketball subset of the MultiSports dataset, focusing on basketball-related scenarios. It contains numerous complex interaction scenes between people and between people and objects, with a larger number of instances. The dataset is modified by removing six videos with excessive repetitive actions that cannot be distinguished by natural language, resulting in a total of 520 videos, with 374 in the training set and 146 in the validation set. Unlike other STVG datasets, each video in SportsGrounding contains multiple captions describing different target persons. Key statistics of the dataset include 4243 video-text pairs, an average video duration of 19.70 seconds, an average tube duration of 1.49 seconds, and an average description length of 16.89 words. Unique characteristics of SportsGrounding include very short durations for some instances, more people appearing in the videos, and more complex interactions between people, with many descriptions requiring certain reasoning and scene information modeling.