README.md

metadata

license: other
task_categories:
  - video-classification
tags:
  - camera-movement
  - video
pretty_name: CamChoreo
extra_gated_prompt: >-
  You agree to use this dataset for non-commercial research use only. The videos
  originate from YouTube and remain under their original owners' copyright; no
  redistribution rights are granted.
extra_gated_fields:
  Name: text
  Title:
    type: select
    options:
      - Prof.
      - Dr. / PhD
      - Mr.
      - Ms.
      - Miss
      - label: Other
        value: other
  Affiliation / Organization: text
  Country: country
  Email: text
  I agree to use this dataset for non-commercial use only: checkbox

CamChoreo

Temporally grounded compositional camera motion benchmark (4229 clips).

Corresponding paper: Temporally Grounded Compositional Camera Motion Understanding via Geometric Knowledge Distillation.

Contents

• videos.zip — 4229 video clips. Unzip to get videos/<video_id>.mp4.
• annotations.jsonl — one record per clip:
  • video_id
  • local_path — ./videos/<video_id>.mp4
  • is_doubt — true = hard sample, false = normal sample
  • segments — per-segment camera-motion labels with time ranges. Only basic_movement is used; the special_movement field can be ignored.

Usage

unzip videos.zip

Then read annotations.jsonl; each local_path points to the extracted clip.

Terminology note

The label Crane used in the annotations is exactly the same camera movement referred to as Pedestal in our paper. They denote identical camera motion — only the wording differs, with no effect on the meaning.

License

• Annotations (annotations.jsonl): released under CC BY 4.0.
• Videos (videos.zip): sourced from YouTube and remain under their original owners' copyright (mostly Standard YouTube License). Provided for non-commercial research only; no redistribution rights are granted.