Machine learning's prevalence and importance have grown exponentially throughout the past few decades, leading to an explosion in the number of frameworks and SDKs that are available to developers. The OpenVINO Toolkit is Intel's entry into this foray. It is a comprehensive set of tools that enables the quick creation of powerful applications that are able to harness computer vision. It also has builtin hardware accelerators and even dedicated processors (Intel Neural Compute Stick 2) to maximize performance.
Deep Learning Superhero ChallengeThe Deep Learning Superhero Challenge is a contest by Hackster.io that aims to empower developers to build intelligent, computer vision-based projects. Some of the categories for these solutions include education and economic growth, environmental sustainability, manufacturing, and healthcare. All of these sectors rely on smart and scalable technologies to increase productivity and effectiveness.
Intel's Neural Compute Stick platform consists of dedicated machine learning accelerators that can offload much of the work that would normally be done by the CPU and instead do it much more efficiently. They are little USB devices that plug into a PC and can be enabled by either doing nothing (Windows 10) or in just a few lines of commands (Linux).
InstallationTo begin developing your competition entry with OpenVINO, start by downloading the full installation package from the website.
This guide covers installation on Ubuntu, so if you are running a different OS, you can view this guide for more information. Navigate to the directory where you downloaded the.tgz file and run tar -xvzf l_openvino_toolkit_p_<version>.tgz to extract it. Then use the cd command to enter the newly created directory. Find and run sudo ./install.sh to go through the CLI installation steps.
That was for the core parts, so there are still a few dependencies left. All you have to do is navigate to the /opt/intel/openvino/install_dependencies directory and run the following command: sudo -E ./install_openvino_dependencies.sh
The last step is to configure a few environment variables. Run source /opt/intel/openvino/bin/setupvars.sh to set them, but keep in mind that this will only be temporary, as they will be lost once the session is ended. To make them permanent, go edit the ~/.bashrc file with your text editor of choice and add source /opt/intel/openvino/bin/setupvars.sh to the very end. This will be run every time a new session is started. There is also a model optimizer that needs to be configured, so run cd /opt/intel/openvino/deployment_tools/model_optimizer/install_prerequisites followed by sudo ./install_prerequisites.sh to do that.
To verify that everything is installed and configured properly, go to /opt/intel/openvino/deployment_tools/demo and run the Squeezenet Image Classification script: ./demo_squeezenet_download_convert_run.sh. If all goes well, you should see something like this:
The OpenVINO installation comes preloaded with several pre-trained models and samples, which is great when getting started or experimenting. There is also an entire model zoo, similar to Pytorch and Tensorflow, that contains numerous models for nearly any project. There are also many samples that demonstrate the capabilities of OpenVINO, and they can be found here.
You can run the license plate recognition sample by entering ./demo_security_barrier_camera.sh. The script downloads three models and uses the car_1.bmp file as a test. It first identifies the object as a vehicle, then after the license plate is found, the characters are passed into the text recognition model for parsing. The resulting frame shows the bounding boxes and labels for each item.
The field of computer vision and machine learning is ever-expanding and can lead to some very innovative projects that tackle real-world problems. Make sure to join the Deep Learning Superhero Challenge and submit your project.
Comments
Please log in or sign up to comment.