We release SVIRO, a synthetic dataset for sceneries in the passenger\ncompartment of ten different vehicles, in order to analyze machine\nlearning-based approaches for their generalization capacities and reliability\nwhen trained on a limited number of variations (e.g. identical backgrounds and\ntextures, few instances per class). This is in contrast to the intrinsically\nhigh variability of common benchmark datasets, which focus on improving the\nstate-of-the-art of general tasks. Our dataset contains bounding boxes for\nobject detection, instance segmentation masks, keypoints for pose estimation\nand depth images for each synthetic scenery as well as images for each\nindividual seat for classification. The advantage of our use-case is twofold:\nThe proximity to a realistic application to benchmark new approaches under\nnovel circumstances while reducing the complexity to a more tractable\nenvironment, such that applications and theoretical questions can be tested on\na more challenging dataset as toy problems. The data and evaluation server are\navailable under https://sviro.kl.dfki.de.\n