Intel chips are everywhere, and now the processing unit manufacturer is looking to make it easier to turn these...
into deep learning chips for extracting intelligence from video data.
The new Intel toolkit, called the Open Visual Inference & Neural Network Optimization toolkit, or OpenVINO, allows developers to write processes for running deep learning models in chips, rather than requiring data to be sent back to cloud or on-premises data stores. Developers can code in common machine learning languages, including Caffe, MXNet and TensorFlow. With OpenVINO, developers can write processes for Intel's Core chips, integrated graphics units and field-programmable gate arrays.
The new Intel toolkit aims to make it easier to run advanced models on video data in order to extract snippets that contain something of interest to a business. This has been a challenge in security video in retail and other settings, where a massive amount of video data is created, but very little is analyzed.
Intel has some catching up to do in getting its chips into edge devices, where most video data is created. While the company dominates the server market, it has a much smaller presence in mobile devices. Qualcomm, for example, has its chips in a range of smartphones. These processing units are already capable of running advanced deep learning models to power consumer applications, like Facebook. When a user uploads a photo, for example, the application runs a deep learning image recognition model in the chip to automatically suggest people to tag.
Qualcomm is starting to use this technology to make a run at the enterprise market. Last month, the company announced its Vision Intelligence Platform in partnership with Microsoft. The combination of Qualcomm mobile deep learning chips and Azure IoT systems aims to improve video processing in home security products.
Intel is also facing pressure from Nvidia, whose GPU deep learning chips are powering a number of advanced applications. Nvidia already has a smart city product that offers functionality for running deep learning on security camera video data.
The OpenVINO Intel toolkit is already being used by Dahua in its smart city and traffic products, GE in its medical imaging offerings, and Hikvision in its industrial and manufacturing safety products, according to an Intel press release.