ChaLearn Gesture Data 2011 (download)

To cite the data, please use:

"ChaLearn Gesture Dataset (CGD 2011), ChaLearn, California, 2011"

Copyright (c) ChaLearn - 2011

This dataset includes:

  • 50,000 gestures recorded with the Kinect camera, including RGB and depth videos,
  • with image sizes 240 x 320 pixels,
  • at 10 frames per second,
  • recorded by 20 different users,
  • grouped in 500 batches of 100 gestures,
  • each batch including 47 sequences of 1 to 5 gestures drawn from various small gesture vocabularies of 8 to 15 gestures,
  • from over 30 different gesture vocabularies.

The data are available in 2 formats:

A lossy compressed AVI format (total < 5 GB)

devel-1-20 + valid-1-20: ~300 MB

Four other archives devel-xx-yy: ~1 GB each

A quasi-lossless AVI format (total < 30 GB)

devel-1-480 + valid-1-20: 500 archives of ~60 MB each

See also the FINAL EVALUATION DATA

Validation and final data labels

The challenge platforms remain open for submission of post-challenge submissions:

Round1, Round2.

However, to facilitate research, we are also providing now the labels for the validation and final data, which were not available to the challenge participants.

QUASI-LOSSLESS FORMAT

The whole dataset in quasi-lossless format (see the README file) is available on-line at:

LOSSY FORMAT

ETH mirror (Zurich, Switzerland)

Orange mirror (Lannion, France)