Skip to content

Dataset · markov-ai

Videos of professional software usage

markov-ai/computer-use-large

48k screencasts (12.3k hours) for training and evaluating video and robotic models.

What's inside

Dataset of 48,478 screen recordings (~12,300 hours) of professional software usage: AutoCAD, Blender, Excel, Photoshop, Salesforce, etc. Videos are sourced from the internet, trimmed of intros and transitions, with audio removed. Language is English; modalities are tables, text, and video.

What you can do with it

  • Train a video action classifier for software usage
  • Evaluate a computer-use agent on interface understanding
  • Fine-tune LoRA for a video model on screencasts
Tasks
Video classification, Robotics
Languages
English
Size
10K–100K rows
Downloads this month
11.6K
Likes
210

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("markov-ai/computer-use-large")