Google edge TPU - What is edge TPU?And How Does It Work?
Google Coral is the company's new machine learning. It does TensorFlow Lite model inference quickly and efficiently. We take a brief look at the Coral Dev Board, which comes with the google edge tpu chip and is now available online.
edge intelligence
AI is now everywhere, in professional applications as well as consumer ones. With the explosion of connected devices, which comes with an increased demand for privacy protection, low latency, and optimized bandwidth, cloud-trained AI models need to run more often. in peripheral.
google Edge TPU is an ASIC custom-designed by Google to run AI solutions at the network's edge. It delivers high performance with minimal footprint and power consumption and enables the deployment of high-precision AI solutions at the network edge.
Edge TPU
Coral Edge TPU is a general-purpose machine learning framework for edge applications developed by Google. Learning models learned in the cloud may be executed.
It's built on Mendel Linux, Google's version of Debian.
Google Coral is often used for object detection.
If you have a pre-trained machine learning model that recognizes objects in video streams, you can deploy it to the google Edge TPU and feed it data from a local video camera.
The EDGE TPU will begin recognizing things locally, rather than streaming the video to the cloud.
The Coral Edge TPU chip is available in a variety of configurations. You should definitely purchase a separate development board that incorporates the System-on-Module (SoM) and is simple to utilize for development.
You may also purchase a separate TPU accelerator device that connects to a PC via a USB, PCIe, or M.2 port. Separately, a System-on-Module is available for integration into bespoke hardware.
End-to-end AI infrastructure
Edge TPU complements Cloud TPU and Google Cloud services to provide end-to-end hardware and software infrastructure, from the cloud to the network edge, to facilitate the deployment of customers' AI solutions.
High performance with minimal footprint and power consumption
With its performance, small form factor, and low power consumption, Edge TPU makes it possible to deploy high-performance AI applications at the network edge.
Combination of hardware, software, and AI algorithms
Edge TPU is more than a hardware solution: it's a combination of custom hardware, open source software, and advanced AI algorithms for high-performance, easy-to-deploy AI solutions at the network edge.
Wide application scope
Edge TPU can be used with a growing number of industrial applications, such as predictive maintenance, anomaly detection, machine vision, robotics, voice recognition and many more.
Its fields of application are numerous: manufacturing industry, medical sector, commerce, smart spaces, transport, on-site solutions, etc.
In comparison to AWS DeepLens
Google Coral is comparable to AWS DeepLens in many aspects. The biggest distinction for developers is that DeepLens interfaces with the AWS cloud.
You can manage your DeepLens devices and deploy your machine learning models using the AWS Console.
Google Coral, on the other hand, is a stand-alone edge device that does not need a Google Cloud connection. In reality, configuring the development board necessitates executing certain really low-level tasks, such as attaching a USB serial connection and installing software.
DeepLens devices are essentially consumer-grade plastic boxes with fixed video cameras. DeepLens is made so that developers can use it at work instead of putting it into their own products.
In comparison, Google Coral's System-on-Module fits the complete system onto a 4048 mm module. All of the processing units, networking functions, ports, 1GB of RAM, and an 8GB eMMC on which the operating system is loaded are included.
You can design a bespoke hardware solution around the Coral SoM if you wish.
The Coral Development Commission
To get started with Google Coral, you need to invest in a Dev Board, which costs about $150. The board resembles Raspberry Pi devices. Once the board is in place, it only needs power and a WiFi connection to work.
Here are some pointers for the first time you install the board.
-Read the instructions carefully at https://coral.ai/docs/dev-board/get-started/.
They walk you through the process of using the device's three separate USB ports and installing the firmware.
- Mac or Linux computers are OK, but Windows will not function. A bash script is used to install the firmware, and it also needs certain specific serial port drivers. They may work with the Windows Subsystem for Linux, but it is much easier to use a Mac or a Linux PC.
- If the USB port does not seem to be working, make sure you are not using a charge-only USB connection. The virtual serial port device will show up on your computer if you use the correct connection.
- We were unable to use the MDT (Mendel Development Tool). Instead, we had to use the serial port to log in to the Linux machine and manually set up SSH.
- Mendel Linux's default username and password is Mendel/mendel. You may use those credentials to log in through a serial port, but the password does not work over SSH. You must include your public key in. ssh/authorized keys.
- You may set up a WiFi network to eliminate the requirement for an Ethernet cable. This is covered in the getting started guide.
- Once you've got a functioning development board, you may want to check out Model Play (https://model.gravitylink.com/). It's an Android app that lets you send machine learning models from the cloud to the Coral development board.
Before you may connect your smartphone to the server, it must be installed on the Coral development board. You must also know the development board's local IP address on your network.
Implementing Machine Learning Models
Assume you now have a functional Coral development board. You may connect to it via SSH from your PC or the Model Play app on your smartphone.
The Getting Started guide includes instructions for running the built-in sample program edgetpu demo. This program will function without the use of a video camera.
It detects cars in video using a recorded video stream and real-time object recognition.
The result is seen in your web browser.
Through the SSH connection, you can also test out several TensorFlow Lite models.
If you have your own models, visit https://coral.ai/docs/edgetpu/models-intro/to learn how to make them compatible with the Coral Edge TPU.
If you just want to experiment with existing models, the Model Play tool makes it extremely simple.
Choose one of the available models and touch the Free button to save it to your smartphone.
Then press the "Run" button to put it into action.
How to Connect a Video Camera and Sensors
If you purchase the Coral development board, be sure you also purchase the video camera and sensor accessories for an additional $50. You will be able to use your machine learning models for more than just looking at still video data.