Abstract

Rocky-CenterNet is an algorithm based on convolutional networks, aimed at detecting rocks and other ellipse-shaped objects. Instead of representing rocks as bounding boxes or segmentation masks, Rocky-CenterNet models rocks as ellipses. Rocky-CenterNet provides a rich description of the shape of the rocks, while running as fast as object detectors based on bounding boxes. The original implementation of Rocky-CenterNet is made available online. It is built on the original implementation of CenterNet, which is itself implemented in Pytorch. Rocky-CenterNet is useful for automation of heavy-duty machinery in mining related tasks, like autonomous loading of material with LHDs and autonomous rock-breaking hammers.