2021/01/19 by Peer Neubert, Neubert, Peer, Stefan Schubert +1 · 3 citations
Engineering · Computer Science · #Ferroelectric and Negative Capacitance Devices #Cellular Automata and Applications
paper · pdf · doi:10.48550/arxiv.2101.07720
Image and video descriptors are an omnipresent tool in computer vision and\nits application fields like mobile robotics. Many hand-crafted and in\nparticular learned image descriptors are numerical vectors with a potentially\n(very) large number of dimensions. Practical considerations like memory\nconsumption or time for comparisons call for the creation of compact\nrepresentations. In this paper, we use hyperdimensional computing (HDC) as an\napproach to systematically combine information from a set of vectors in a\nsingle vector of the same dimensionality. HDC is a known technique to perform\nsymbolic processing with distributed representation in numerical vectors with\nthousands of dimensions. We present a HDC implementation that is suitable for\nprocessing the output of existing and future (deep-learning based) image\ndescriptors. We discuss how this can be used as a framework to process\ndescriptors together with additional knowledge by simple and fast vector\noperations. A concrete outcome is a novel HDC-based approach to aggregate a set\nof local image descriptors together with their image positions in a single\nholistic descriptor. The comparison to available holistic descriptors and\naggregation methods on a series of standard mobile robotics place recognition\nexperiments shows a 20% improvement in average performance compared to\nrunner-up and 3.6x better worst-case performance.\n