adrianmele/computer-use-large

一个大规模屏幕录制视频数据集，包含48,478个专业软件使用视频（约12,300小时），来源自互联网。所有视频均经过修剪，去除了非屏幕录制内容（如介绍、结尾、人物讲话、过渡等）和音频。数据集按软件类别（如AutoCAD、Blender、Excel、Photoshop、Salesforce和VS Code）组织，每个类别包含不同数量的视频和时长。每个视频文件夹包含一个metadata.jsonl文件，记录文件名、类别、修剪后的时长和连续屏幕录制段数等信息。数据集旨在训练和评估计算机使用代理，即通过GUI操作（点击、输入、滚动）与桌面软件交互的模型。

A large-scale dataset of 48,478 screen recording videos (~12,300 hours) of professional software being used, sourced from the internet. All videos have been trimmed to remove non-screen-recording content (intros, outros, talking heads, transitions) and audio has been stripped. The dataset is organized by software categories (e.g., AutoCAD, Blender, Excel, Photoshop, Salesforce, and VS Code), each containing a varying number of videos and durations. Each video folder includes a metadata.jsonl file recording the file name, category, trimmed duration, and number of contiguous screen recording segments. The dataset is designed for training and evaluating computer use agents — models that interact with desktop software through GUI actions (clicking, typing, scrolling).