markov-ai/sample-100-hours
收藏资源简介:
Markov样本数据集—100小时(CAD与设计)是一个包含101小时Windows和macOS屏幕录制的数据集,附带同步的键盘/鼠标事件数据和逐步叙述,专注于计算机辅助设计(CAD)和设计工作流程。每个工作流程包含多个文件:屏幕录制视频(clip.mp4)、原始事件数据(events.json)、帧对齐事件数据(frame_events.json)、元数据(metadata.json)以及时间对齐的叙述(narration.json),其中叙述包括每个步骤的动作、意图和应用信息。数据集总计133个工作流程,总时长101.3小时,包含187,838个注释和27.1百万个事件。统计信息显示,工作流程时长中位数为31.1分钟,类别分布以CAD为主(占81.6%),平台以Windows为主(占96.7%),应用包括AutoCAD、SketchUp、Blender等。叙述模式采用JSON格式,提供详细的动作、意图和应用上下文,用于分析和理解用户交互行为。
The Markov Sample Dataset — 100 Hours (CAD & Design) consists of 101 hours of Windows and macOS screen recordings with synchronized keystroke/mouse data and step-by-step narrations, spanning CAD and design workflows. Each workflow includes: clip.mp4 (screen recording at 30fps CFR, H.264), events.json (raw mouse/keyboard/app events in JSONL), frame_events.json (frame-aligned mouse/keyboard/app events), metadata.json (workflow metadata such as category, title, description, tags, duration), and narration.json (time-aligned narration with action, intent, and app per step). The dataset contains 133 workflows, totaling 101.3 hours, with 187,838 annotations and 27.1M events. Statistics include workflow duration percentiles (e.g., median 31.1 min), category distribution (CAD 81.6%, Design 18.4%), platform distribution (Windows 96.7%, macOS 3.3%), and app distribution by hours (e.g., AutoCAD 22.5%, SketchUp 15.2%). The narration schema provides a flat list of time-ordered annotations with start/end times, action, intent, app, and details, plus a workflow summary.




