KITTI-depth contains more than 93,000 depth maps and the corresponding raw LiDaR scans and RGB images. Given the large amount of training data, this dataset should allow training complex deep learning models for the tasks of depth completion and single image depth prediction. In addition, the dataset provides manually selected images with unpublished depth maps as a benchmark for these two challenging tasks. The dataset has 86K training images, 7K validation images, and 1K test images on the benchmark server without ground truth.