Skip to content

anibali/h36m-fetch

Repository files navigation

Human3.6M dataset fetcher

Human3.6M is a 3D human pose dataset containing 3.6 million human poses and corresponding images. The scripts in this repository make it easy to download, extract, and preprocess the images and annotations from Human3.6M.

Please do not ask me for a copy of the Human3.6M dataset. I do not own the data, nor do I have permission to redistribute it. Please visit http://vision.imar.ro/human3.6m/ in order to request access and contact the maintainers of the dataset.

Requirements

  • Python 3
  • axel
  • CDF
  • ffmpeg 3.2.4

Alternatively, a Dockerfile is provided which has all of the requirements set up. You can use it to run scripts like so:

$ docker-compose run --rm --user="$(id -u):$(id -g)" main python3 <script>

Usage

  1. Firstly, you will need to create an account at http://vision.imar.ro/human3.6m/ to gain access to the dataset.
  2. Once your account has been approved, log in and inspect your cookies to find your PHPSESSID.
  3. Copy the configuration file config.ini.example to config.ini and fill in your PHPSESSID.
  4. Use the download_all.py script to download the dataset, extract_all.py to extract the downloaded archives, and process_all.py to preprocess the dataset into an easier to use format.

Frame sampling

Not all frames are selected during the preprocessing step. We assume that the data will be used in the Protocol #2 setup (see "Compositional Human Pose Regression"), so for subjects S9 and S11 every 64th frame is used. For the training subjects (S1, S5, S6, S7, and S8), only "interesting" frames are used. That is, near-duplicate frames during periods of low movement are skipped.

You can edit select_frame_indices_to_include() in process_all.py to change this behaviour.

License

The code in this repository is licensed under the terms of the Apache License, Version 2.0.

Please read the license agreement for the Human3.6M dataset itself, which specifies citations you must make when using the data in your own research. The file metadata.xml is directly copied from the "Visualisation and large scale prediction software" bundle from the Human3.6M website, and is subject to the same license agreement.