MovieGraphs Dataset

Dataset release for our paper published at CVPR 2018.

📜 MovieGraphs: Towards Understanding Human-Centric Situations from Videos

🧑‍💻 Project page

Release updates

• March 2019 (Original release with python 2.7; shared over emails)
• June 2025 (python 3.10 support)

Files concerning movies

• dvds.txt: links to Amazon DVDs
• movies_list.txt: just a list of movies
• split.json: train/validation/test splits

Nomenclature of video scenes

• xxx: scene id, starting from 1
• yyyy: start shot for scene xxx starting from 1
• zzzz: end shot for scene xxx
• moviegraphs_startend_frames: first and last frame of every movie scene clip (tarball, please expand)
• shot boundaries (videvents) and scene GT boundaries

[June 2025] Updated release

• py3loader_new
  • all_movies.pkl: main pickle file containing the parsed graph annotations.
  • GraphClasses.py: main ClipGraph and MovieGraph python classes. There are many helper functions here, please read this.
  • startup.py: a little test to make sure everything is setup correctly. This should run without errors.
  • tutorial.ipynb: a small notebook that tests functionality of some parts of GraphClasses.py. If you run all cells, it should create a pdf file with all graphs of one movie.
• nx_code: copies and updates essential files from networkx=1.10 that were required for GraphClasses.py to work properly with python 3.

Videos

⚠️🎞️ Full videos are not shared and should be procured using the DVDs listed in dvds.txt. If you intend to use the dataset for non-commercial and academic research purposes, a 1 fps version can be downloaded here.

Subtitles can be downloaded here.

[Deprecated] Original release (shared over emails)

Worked with py2.7 and networkx1.10

• py3loader
  • 2017-11-02-51-7637_py3.pkl: main pickle file containing the parsed graph annotations
  • GraphClasses.py: main ClipGraph and MovieGraph python classes
  • startup.py: a little test file to make sure things are working as expected

Acknowledgement

Thanks to Lakshmipathi Balaji for helping with the new release!