Ground-truth datasets: (a) rendering of the mesh and sensors positions of the “bunny” dataset. 36 sensors in two elevations are placed in the scene. Ground truth for each sensor is computed directly from the mesh. Random undersampling and noisifying of the bunny object (without plate) and outlier injection are used at different levels to generate depth maps for quantitative analysis. (b) Ground-truth laser scan of the “fountain” dataset and the related 11 cameras (registered to the scan).