Opening Doors to Computer Vision & Deep Learning for Developers

OpenVino Toolkit Intro image

At the Embedded Vision Summit in Santa Clara this week, I’m excited to meet with fellow innovators and leaders – all working toward the space of bringing data visualization to life. 

Computer vision is the science and technology used by machines to see as a human eye would see. This includes all the hardware and software for capturing, processing, analyzing and understanding digital images.  

We have cameras in our smart phones, our cars and homes, and around our cities. And now computer vision is spreading across new industries such as sports, retail, healthcare, robotics, aero-journalism, agriculture, transportation – and more. Despite the differences in applications and business models across markets, the core technologies to execute computer vision – or even transform it into artificial intelligence (AI) - are the same.

Today, developers and data scientists no longer have to be domain experts working at major research centers to bring computer vision and deep learning to life in applications. Every developer with the interest and passion in computer technology has access to a field of open source tools, libraries, frameworks and deep learning models to build their expertise to make computer vision a reality. It’s now an open door for mass developers to put vision inside applications.

What has changed to make computer vision so accessible? We have affordable hardware now - overcoming previous cost barriers available only to the likes of researchers and universities. Heterogeneous hardware and special accelerators are also part of the mix so developers have choices to pick the right performance, cost, flexibility, and scalability for their specific solution needs. 

And thanks to Moore’s Law,  - the evolution of computing and technology advancements bring video and image processing, analytics, streaming, and understanding to real-time – adding significant inflection points to efficiency and breaking down barriers to bandwidth constraints. 

You may have seen some of the press last week, but if you haven’t – Intel announced its portfolio of end-to-end vision solution products. As part this offering, the OpenVINO™ toolkit (Open Visual Inference & Neural Network Optimization), is a free software package that helps to fast-track the development of high-performance computer vision and deep learning inference into a wide range of applications, from edge to cloud. 

The OpenVINO toolkit supports a broad range of functional use cases, including on premise security, object recognition, face detection and more. It delivers optimized libraries and trained models to accelerate your applications as well as your development cycle. You write your code or algorithms once and can deploy easily to multiple types of Intel® platforms and hardware accelerators: CPUs, CPUs with integrated graphics (Intel® Processor Graphics), FPGAs (field programmable gate arrays), and Movidius™ vision processing units (VPUs).

To make things easier for developers, the OpenVINO toolkit is based on common development standards such as OpenCL™, OpenCV* and OpenVX*, has a common API, includes many code samples and 10 pre-trained models, and supports 100+ open source and custom models. Check it out.

Overall, it’s an incredible time for computer vision, and an amazing opportunity for developers to innovate.

Visit the Intel Developer Zone for your free download of the OpenVINO toolkit.

About the Author

Charlotte DrydenCharlotte Dryden leads Intel’s Visual Computing Developer Solutions team, which defines and develops software products and tools that enable visual computing solutions.  

Charlotte has 20 years of technical and international business experience in the mobile communications, semiconductor and wireless industries, with a focus on designing software solutions to differentiate hardware. Find Charlotte on Linked-In, and follow her team’s efforts @Inteldevtools.


What’s inside the OpenVINO™ toolkit:
OpenVino toolkit
  • The Intel® Deep Learning Deployment Toolkit to deploy pre-trained models from popular frameworks such as Caffe*, TensorFlow* and MxNet* through a high-level C++ inference engine API integrated with application logic. It includes:
    • A Model Optimizer: a Python-based command line tool imports pre-trained models and converts them to a unified intermediate representation file, and it optimizes topologies through node merging, and horizontal fusion, to eliminate the need for batch normalization and quantization 
    • The Inference Engine: a simple and unified API to deliver inference solutions on the platform of your choice: CPU, CPUs with integrated graphics (Intel® Processor Graphics), VPU, or FPGA 
    • 10 pre-trained models: already optimized and ready to use
  • Optimized computer vision libraries for OpenCV, OpenVX and Photography Vision for Intel CPUs and Intel Processor Graphics 
  • Components to increase performance of Intel Processor Graphics for Linux*, including the Intel® Media SDK open source version and OpenCL graphics drivers and runtime
  • FPGA Runtime Environment (from the Intel FPGA SDK for OpenCL) and bitstreams for Linux FPGA
In addition, you can harness even more performance when using the toolkit with other tools, such as:
  • Intel Media SDK to accelerate video and image encode/decode
  • Intel® SDK for OpenCL Applications for CPU/CPU with integrated graphics workload balancing
  • Intel® System Studio 2018 to optimize system bring-up and IOT device application performance
For more complete information about compiler optimizations, see our Optimization Notice.