Intel is unveiling OpenVINO, a toolkit designed to easily bring computer vision and deep-learning inference to vision applications at the edge.
The OpenVINO (Open Visual Inference & Neural network Optimization) toolkit enables developers to build and train AI models in the cloud—on popular frameworks such as TensorFlow, MXNet, and Caffe—and deploy them across a broad range of products. It takes advantage of Intel’s investments in different AI accelerator technologies, including CPUs, field programmable gate arrays (FPGAs), and Movidius vision processing units (VPUs).
OpenVINO provides a set of optimization capabilities and a runtime engine that allows developers to run their model on the architecture that best suits their needs, whether it’s a highly-tuned FPGA, an efficient VPU, or another choice. For instance, in a retail setting, a developer might want to deploy computer vision capabilities within a range of edge applications, including the point of sale, within digital signage and within security cameras.
Intel says it’s seeing significant growth in IoT markets worldwide, in part because of a significant increase in vision applications leveraging AI. Businesses need intelligence at the edge for a number of reasons, Intel’s Adam Burns said: They may want to store less data longterm, they may be driven by bandwidth constraints, or they may need to make immediate decisions based on the data collected. Intel’s IoT business in Q1 grew at a rate of 17% year over year to deliver $840 million in revenue.
OpenVINO should be applicable across a broad range of markets, including industrial, retail, energy, and medical, Burns said. It’s already being used for a variety of use cases—for instance, Dahua is using it for smart city and traffic solutions, and GE Healthcare is using it for medical imaging. Other companies using it include AgentVi, Amazon Web Services, Dell, and Honeywell.
