MODS—A USV-Oriented Object Detection and Obstacle Segmentation Benchmark

Small-sized unmanned surface vehicles (USV) are coastal water devices with a broad range of applications such as environmental control and surveillance. A crucial capability for autonomous operation is obstacle detection timely reaction collision avoidance, which has been recently explored in the context camera-based visual scene interpretation. Owing to curated datasets, substantial advances interpretation have made related field ground vehicles. However, current maritime datasets do not adequately capture complexity real-world USV scenes evaluation protocols standardised, makes cross-paper comparison different methods difficult hinders progress. To address these issues, we introduce new benchmark MODS, considers two major perception tasks: object more general segmentation. We present diverse dataset containing approximately 81k stereo images synchronized an on-board IMU, over 60k objects annotated. propose segmentation performance protocol that reflects accuracy way meaningful practical navigation. Nineteen recent state-of-the-art evaluated using proposed protocol, creating facilitate development field. The dataset, well routines, publicly available at vicos.si/resources.