Here are a few cool things that came up at CVPR. Today KIT released a benchmark data set for autonomous vehicles. KIT has spent a small fortune outfitting a vehicle with a Velodyne LIDAR, a stereo camera rig, GPS, INS, and all of the other goodies you would expect a DARPA urban challenge vehicle to have. KIT drove the vehicle around a city and recorded six hours of real world data. KIT then painstakingly rectified everything together and paid someone to mechanically segment and classify the data in the scenes (i.e. all pedestrians and vehicles have 3D boxes around them and are labeled in the data). The data is also registered to open street map data. This means that the world now has open source, real-world data for autonomous vehicle navigation. Since KIT provides benchmark along with the data it should be trivial to use the data and compare how your algorithms perform. This work will really serve to drive competition in the field.

Tomorrow at CVPR Kitware is hosting a Python for Computer Vision workshop. Kitware provides open source python tools for computer vision, and they have opened up the materials. You can find them here. I will report more information tomorrow after the workshop