Getting started with the Google Coral Edge TPU USB Accelerator: running video image recognition

[@WFB-ng]

Excellent ! can it detect patterns like aruco markers or qr code pattern ?

[@ExpatProfessor]

You do not need any Coral Edge TPU to detect these things you've mentioned. An USB camera and any machine, even Raspberry Pi is capable of doing it with the right Python script and libraries.

[@WFB-ng]

@@ExpatProfessor Hi, I have done that alreday; kzbin.info/www/bejne/hJOkqmSLa55npMU I was thinking if google coral usb accelerator could enhance the performance of my system in terms of video resolution which is currently 640x480 to may be 1080p or atleast 720p and fps which is currently 15-20fps to minimum 30 fps.

[@sosome1]

Thank you for posting this great tutorial. I was looking for it for quite a while. I am looking for a board to place inside my car. I want to use realtime object detection to recognise whenever a certain type of car is passing by in front of my car. And I want the board to notify me by 4g/lte sms( no wifi connection so i have to add a gsm/lte/4g module I am looking at 2 boards: 1. Raspberry pi 5 8Gb with Google Coral USB Accelerator 2. Google Coral Dev Board 4Gb Which one is the best option?

[@ExpatProfessor]

Hello, either one is fine as long as the RB Pi is hooked to google coral edge TPU. It can even be RB Pi 4. I tested RB Pi 5 and the efficiency in object detection is minuscule. Great idea to add text sms notification. It did that as my side project and I sent notifications by sms via twilio API with the use of LTE from a hot spot. I did not record any of that however this question gives me the motivation to repeat and post to YT. So the Edge TPU dev board or RB pi 4/5 with edge TPU is all fine!

[@sosome1]

@@ExpatProfessor thank you. looking forward to your future videos

[@Burgerhs]

I tried the coral usb on windows and Ubuntu nothing 😢 I'm returning mine I've been trying for over a week now

[@ExpatProfessor]

What a pity, perhaps you could give the Coral Dev Board a try instead...

[@Burgerhs]

What about fedora 40​@MachineLearningSebastian

[@evgenybenediktov8964]

The problem with all these TPUs is that you can only run the models from the examples. So far, I didn't manage to find a tutorial on how to design and train on an Nvidia card a CNN model that will run on Coral or any other Pi TPU

[@rbrisita]

All NPUs will have their own formats. The approach is to convert existing models to the target NPUs format; number precision and matrix dimensions.

[@saramasood4675]

Why its not working anymore

[@ExpatProfessor]

Do you have any details?

[@jxzmh]

Not working because i have Python 3.11 and I need 3.6-3.9 but doesn’t work to downgrade it

[@ExpatProfessor]

Apparently, the Python version manager tool "pyenv" is the solution. With this library, you can choose the desired version of python maxat-akbanov.com/how-to-install-pyenv-python-version-manager-on-ubuntu-2004

[@Warezzfan]

@@ExpatProfessor Python pyenv also doesn't work. Coral libraries disregard pyenv and only install for older version of Linux which come with python version less than 3.9

[@rbrisita]

@@Warezzfan pyenv local 3.9

[@Warezzfan]

This ONLY works for python version 3.6 - 3.9. NO the "pyenv" doesn't solve the problem. If your OS natively runs with python 3.6 to 3.9 then opt USB Coral, otherwise for newer OS better look for something else. Coral dev team already made enough money out of this product, and are NOT willing to provide support to fix any issues.