README.md

metadata

license: cc-by-4.0
pretty_name: video2sim captures
language:
  - en
task_categories:
  - video-classification
  - image-to-3d
tags:
  - video
  - 3d-reconstruction
  - robotics
  - simulation
  - benchmark
size_categories:
  - n<1K

video2sim captures

Short handheld videos of workspaces, captured as input for the video2sim pipeline (video → reconstruction → segmentation → shape → physics → simulation assets).

Contents

• scenes/s01–s24/: 24 scenes, each with:
  • video_a.mp4 — orbit around the table
  • video_b.mp4 — rough side view
  • alt_*/ — extra takes (alt_lighting, alt_path, alt_closeup, glass_bottle_close)
• scenes/thumbnails/, scenes/thumbnails_lowres/: preview images (1920×1080 and 640×360)

Clips are .mp4, 5–65 seconds each. Scenes s06, s14 and s24 are video_b-only (plus two alt_path takes each).

Capture protocol

• Device: iPhone 14
• Resolution: 1920×1080
• Frame rate: 30 fps
• Camera motion: VideoA is an orbit around the table, VideoB a rough side view
• Lighting: mostly diffuse room light (optimal conditions); some scenes use natural light

Privacy and consent

Scenes are clean for the most part. One background document still needs a double-check before public release.

Intended use

Research and benchmarking of video-to-simulation pipelines.

Citation

@dataset{korbmacher2026video2sim,
  author    = {Korbmacher, Tom Korbmacher, Tom},
  title     = {video2sim captures},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/TomKrb/video2sim-captures}
}