Distributed Computing

DeMR Network powers reconstruction tasks and real-time service for MR spatial scenesvia its powerful distributed computing capability. Each Computing Node runs components of the DeMR cloud service. The components include a deep neural network-based image feature extractor, image matcher, structure from motion component. The distributed cluster splits a reconstruction task into several sub-tasks and distributes the sub-tasks to different Computing Nodes. The tasks are supervised and managed by decentralized nodes. On each Computing Node, the reconstruction algorithm can take advantage of the heterogeneous computing capability of multi-core GPU, GPU, and NPU (if available). A high-quality MR spatial scene is finalized after tens of steps including several rounds of refinement.

The distributed spatial scene reconstruction procedure ensures the high efficiency of building. It takes only a few hours to complete the whole process of rebuilding a block while guaranteeing precision. The cluster is decentralized and dynamically scalable, multiple DeMR reconstruction tasks can run simultaneously, Computing Node resource is utilized maximumly, and the computing cost is reduced significantly.

DeMR Network contains loads of advanced technologies including heterogeneous computing of multi-core GPU, GPU, and NPU, large-scale Structure from Motion (SfM), semantic SLAM, 6DoF tracking, and massive multiplayer online interaction. DeMR provides open API and development tools for global developers to develop abundant MR DAPPs, building a wonderful future spatial Internet.