Image dataset.

We have collected a people and non-people image dataset.

• Negative images have been chosen from the LabelMe dataset*.

  LabelMe is a database and an online annotation tool that allows the sharing of images and annotations. The online tool rovides functionalities such as drawing polygons, querying images, and browsing the database.

  We have selected images with objects and scenarios without people. Each image has been cropped in small pieces in order to obtain a huge number of different negative images and has been normalized (four diferent body part sizes and gray scaled).

• Positive images hasve been chosen from the INRIA dataset**.

  INRIA contains images of humans cropped from a varied set of personal photos, the people are usually standing, but appear in any orientation and against a wide variety of background image including crowds.

  People body blobs have been extracted, normalized (58x24 pixels and gray scaled) and have been segmented in body parts.

  

Finally our image dataset stores 3542 positive images and more than 40000 negative images.

* LabelMe dataset

B.C. Russell and A. Torralba and K.P. Murphy and W.T. Freeman, LabelMe: A Database and Web-Based Tool for Image Annotation, International Journal of Computer Vision (2008), pp. 157-173.

** INRIA dataset

N. Dalal and B. Triggs, Histograms of oriented gradients for human detection, Proc. of CVPR, Vol. 1, 2005, pp. 88-6893.