Custom Vision Identification
The process is very easy as we use the Custom Vision (http://customvision.ai ) service to train an image classifier and then export the finished Model in a number of different forms – CoreML, TensorFlow, ONNX, or dockerfile for IoT Edge/Azure Functions/Azure ML.
So the process is:
- Use CustomVision.ai to build a model
- Export the modal appropriately e.g. TensorFlow
- Use boiler plate IoT Edge module code and swap in new modal files.
- Custom test data source as appropriate and execute against models
For an end to end example see https://www.hackster.io/glovebox/image-recognition-with-azure-iot-edge-and-cognitive-services-a2b939
What can this run on?
The evolution of this is to exploit the Myriad processors from Intel by running a model – such as those provided by Intel’s OpenVino – on the Myriad processor under the control of a Pi. This has enabled some significant CNN models to be executed.
We are containerising our cognitive services (https://docs.microsoft.com/en-us/azure/cognitive-services/cognitive-services-container-support ), currently available are CV – OCR, Face, LUIS, Text Analytics – key phrase extraction, Text Analytics – Language Detection, Text Analytics – Sentiment analysis.
Voice recognition with key phrase activation is an interesting topic. There are of course OS features for Linux and Windows which provider this at low edge of devices – for example Raspberry Pi. Windows IoT Core has speech recognition for a number of languages and there are a number of examples of simple Voice Assistants, such an example is - https://wotudonet.wordpress.com/2016/08/04/project-patrick-build-your-own-personal-assistant/
There are also specialist chips which provide local voice processing, specifically activation word recognition, such as the XMOS chip used in the Respeaker.Io expansion board. https://www.xmos.com/
Respeaker even have a workshop for the end to end architecture of a smart speaker: https://respeaker.io/make_a_smart_speaker/
Embedded Learning Library
ELL provides models for Resource Constrained Platforms and single-board computers. https://github.com/Microsoft/ELL
The Embedded Learning Library (ELL) allows you to design and deploy intelligent machine-learned models onto resource constrained platforms and small single-board computers, like Raspberry Pi, Arduino, and micro:bit. The deployed models run locally, without requiring a network connection and without relying on servers in the cloud. ELL is an early preview of the embedded AI and machine learning technologies developed at Microsoft Research.
Go to our website for tutorials, instructions, and a gallery of pretrained ELL models for use in your projects.
This post first appeared on MSDN Blogs | Get The Latest Information, Insights, Announcements, And News From Microsoft Experts And Developers In The MSDN Blogs., please read the originial post: here