A large scale 3D reconstruction based on the KITTI dataset. While the car is moving, the top-mounted camera produces an RGB stream semantically segmented in real time by means a Deep Neural Network. Using the camera 6-DoF pose tracked during the mission and 3D information from a Lidar we build a global 3D Voxel Grid [1] representation of the environment which incorporates semantic information. In this representation, each voxel is painted with the color associated with its class.