Contextual and video data needed for Subjective Cyclist Experience modelling

This dataset contains contextual GIS data needed for the analysis, manual, and LLM-based video data annotations (excluding videos themselves). The dataset includes four main components: Video contextual data – open-source GIS layers with metadata that define video context. Manually-labelled video data – human-annotated labels generated through manual annotation, providing curated and expert-informed interpretations of video content. LLM-labeled video data – labels automatically generated using Large Language Models (LLMs), intended for comparison with human annotations and exploration of automated labeling approaches. Video ground truth – finalized reference labels used for evaluation and benchmarking, derived from LLM annotations, manual annotations, and geospatial operations. The dataset is suitable for research in video analysis, machine learning, multimodal learning, and annotation quality comparison. The dataset does not include videos themselves.本数据集包含用于分析、人工标注及基于大语言模型（Large Language Model，LLM）的视频数据标注所需的上下文地理信息系统（Geographic Information System，GIS）数据（不含视频本体）。本数据集包含四大核心组成部分： 1. 视频上下文数据：搭载元数据的开源GIS图层，用于定义视频相关上下文； 2. 人工标注视频数据：通过人工标注流程生成的人类标注标签，提供经筛选且经专家研判的视频内容解读结果； 3. 大语言模型标注视频数据：使用大语言模型自动生成的标签，用于与人工标注结果对比并探索自动化标注方案； 4. 视频基准真值：用于评估与基准测试的最终参考标签，源自大语言模型标注、人工标注及地理空间运算。 本数据集适用于视频分析、机器学习、多模态学习以及标注质量对比等方向的研究。本数据集不包含视频本体。