APR May Jun
Previous capture 11 Next capture
2015 2016 2017
13 captures
14 May 15 - 11 May 16
sparklines
Close Help

FLIC-motion Dataset

Overview:

The FLIC-motion dataset is a set of 30 Hollywood movies from which the original FLIC dataset of 5003 labeled RGB images were collected along with the information about the location of all these 5003 images in the 30 movies. It was first introduced as part of our ACCV paper: MoDeep: A Deep Learning Framework Using Motion Features for Human Pose Estimation, to facilitate the use additional motion-features for the task of pose-estimtion.

For motivation and details on how the dataset was created, please refer to our paper.

To Download:

  • Download our examples.mat
  • Download our movie files (28GB) FLICMOVIES.tar.gz
  • Note: This has only been tested on Windows. Load this downloadedexamples.mat into Matlab.
  • Decompress FLICMOVIES.tar.gz
  • The struct examples is the exactly the same as the one from the original FLIC dataset, except that the values of field examples.currframe has been modified to match the correct frame number in our video files.
  • To locate example(3533).filepath in the video, do:
    idx = 3533;
    movieName = ['FLICMOVIES/' examples(idx).moviename '.m4v'];
    vidObj = VideoReader(movieName);
    im_from_video = read(vidObj, examples(idx).currframe);
    imshow(im_from_video);
  • For optical flow, we refer you to DeepFlow.

Citing our paper (and this dataset):

@article{jain14accv,
  author = {Arjun Jain and Jonathan Tompson and Yann Lecun and Christoph Bregler}
  title = {MoDeep: A Deep Learning Framework Using Motion Features for Human Pose Estimation,
  journal = {ACCV},
  year = {2014},
}

Please make sure you also cite Sapp et al. for first introducing the FLIC dataset!

Publication

ACCV'14 paper