USC Pedestrian Detection Test Set


The "USC pedestrian set A" is collected from the Internet to evaluate frontal/rear view walking/standing human detection algorithm without inter-human occlusion. This set contains 205 images, with 313 humans. The groundtruth is provided in XML format.

The "USC pedestrian set B" is collected from the CAVIAR video corpus to evaluate frontal/rear view walking/standing human detection algorithm in presence of partially inter-human occlusion. This set contains 54 images, with 271 humans. The groundtruth is provided in XML format.

The "USC pedestrian set C" is collected from the Internet to evaluate multi-view walking/standing human detection algorithm without inter-human occlusion. This set contains 100 images, with 232 humans. The groundtruth is provided in XML format.

The three sets are free to use for research purposes. If your publication includes any experimental results obtained on any images from set A/B, please acknowlege the use of the "USC pedestrian set A/B" and reference the following paper:
Bo Wu, and Ram Nevatia. Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors. ICCV 2005.

If your publication includes any experimental results obtained on any images from set C, please acknowlege the use of the "USC pedestrian set C" and reference the following paper:
Bo Wu, and Ram Nevatia. Cluster Boosted Tree Classifier for Multi-View, Multi-Pose Object Detection. ICCV 2007.

By pressing the "Accept" button you are accepting all the above terms and will be forwarded to the downloading page.