Multimedia Laboratory

Datasets / CUHK Occlusion Dataset

Download

CUHK occlusion data set is for research on activity analysis and crowded scenes. This dataset contains 1063 images with occluded pedestrians from the datasets of Caltech [1], ETHZ [2], TUD-Brussels [3], INRIA [4], Caviar[5] and images collected by us. It is divided into 10 clips and can be downloaded from the following links.

In order to evaluate the performance of human detection on this data set, ground truth of pedestrians all images is manually labeled. It can be downloaded below. A readme file provides the instructions of how to use it.

Ground truth of pedestrians

In order to open the sequence and label, please directly run the vbbLabeler.m in the Caltech Toolbox and open the .seq and .vbb above. More tools of Caltech are provided here.

References

  1. A Discriminative Deep Model for Pedestrian Detection with Occlusion Handling
    W. Ouyang and X. Wang
    IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 2012

Literature

  1. Pedestrian detection: An evaluation of the state of the art
    P. Doll´ar, C. Wojek, B. Schiele, and P. Perona
    TPAMI, 2011
  2. Depth and appearance for mobile scene analysis
    A. Ess, B. Leibe, and L. V. Gool
    ICCV, 2007
  3. Multi-cue onboard pedestrian detection
    C.Wojek, S.Walk, and B. Schiele
    CVPR, 2009
  4. Histograms of oriented gradients for human detection
    N. Dalal and B. Triggs
    IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2005
  5. Caviar: homepages.inf.ed.ac.uk/rbf/caviar/
    R. Fisher


Back to Top

Share This Page

© 2013-2019Multimedia Laboratory

The Chinese University of Hong Kong 香港中文大学

Site Credits