In this video, we walk you through various components

of the OpenVino toolkit.

Deep learning inference occurs at the edge.

which is optimized for power and performance.

Developers can learn how to write optimized code

for different hardware accelerators.

However, achieving proficiency will take a long time.

To accelerate deep learning application development,

Intel has introduced the OpenVino Toolkit,

or the Open Virtual Inference and Neural Network Optimization

Toolkit.

This convolutional neural network based toolkit

is designed to increase performance, reduce power,

maximize hardware utilization, and reduce time to market

for computer vision solutions.

Now, I will dive into different components of the OpenVino

toolkit.

It includes tools for deep learning,

as well as for traditional computer vision.

Let's start with the deep learning deployment toolkit.

It is a cross-platform tool to accelerate deep

learning inference performance.

It includes model optimizer and inference engine.

The model optimizer takes pre-trained models

from various frameworks, such as Caffe, Tensorflow, MXnet

and converts them into an intermediate representation

that is IR files.

The inference engine takes these IR files as input,

and then using the unified API, it deploys the computer vision

application on different platforms,

like CPU, integrated GPU, Movidius compute

stick, or FPGs without making any code changes.

The OpenVino Toolkit also includes OpenCV and OpenVX

for traditional computer vision.

It includes optimized OpenCV libraries

for Intel architecture.

Additionally, a new Intel photography vision library

with face detection, recognition, blink detection,

and smile detection is also added.

The OpenVino Toolkit also includes

the media SDK to support hardware optimized

encode, decode, and pre-processing of the input

image or video.

It supports OpenCL to run the application

on various platforms and add custom layers.

In addition to all these components,

this package also gives you the Intel FPGA Deep Learning

Acceleration suite, which includes

precompiled bitstreams and the Intel FPGA SDK for OpenCL.

Let's now look at how you can utilize OpenVino Toolkit

components to develop an optimized computer vision

application.

First, the OpenVino Toolkit takes three trained models

from frameworks such as Caffe, Tensorflow, and MXnet.

Model optimizer imports these models

and converts them into intermediate formats

to be processed by the inference engine.

Then, based on your choice, the inference engine API

runs the application on various hardware types,

such as CPU, integrated GPU, Movidius Neural Compute Stick,

or FPGA.

OpenVino also supports the heterogeneous architecture

with fallback to a different device for custom

or unsupported layers.

It also features an asynchronous execution,

which allows you to perform other tasks while the hardware

accelerator is crunching the current frame.

The OpenVino Toolkit comes with included samples that

showcase basic functionality.

These samples use Intel pre-trained detection

and recognition models for various tasks,

from age and gender recognition to multiple object detection,

and even headpose estimates.

We also offer a model downloader to access

several public pre-trained inference models, such as SSD,

VGG, DenseNet, and SqueezeNet.

To learn how to install the OpenVino Toolkit,

follow the link provided to the installation page.

The product website has a growing set of resources

to help you ramp up faster, including documentation,

training, and a support forum.

Thanks for watching.

In the next video, we cover the model optimizer in detail.