Semantic Monocular Depth Estimation Based on Artificial Intelligence

Depth estimation provides essential information to perform autonomous driving and driver assistance. A promising line of work consists introducing additional semantic about the traffic scene when training CNNs for depth estimation. In practice, this means that data used CNN is complemented with images having pixel-wise labels where same raw associated both types ground truth, i.e., labels. The main contribution paper show hard constraint can be circumvented, we train by leveraging coming from heterogeneous datasets. order illustrate benefits our approach, combine KITTI Cityscapes segmentation datasets, outperforming state-of-the-art results on monocular