This guide provides step-by-step instructions to set up the Google Coral USB Accelerator on Windows 11 with WSL2 using Ubuntu 24.04 LTS. Due to some outdated documentation on the official website, this guide includes the necessary workarounds to ensure proper installation.
Table of Contents
- Introduction
- Prerequisites
- Step-by-Step Guide
- Run a model on the Edge TPU
- Troubleshooting
- WSL with Web Cam Support
The Coral USB Accelerator by Google is a USB device that provides an Edge TPU, a small ASIC designed to accelerate machine learning inference at the edge. However, setting up the Coral USB Accelerator on WSL2 can present challenges due to outdated documentation and Python version issues.
This guide aims to help you successfully set up the Coral USB Accelerator on a system running Windows 11 with WSL2 and Ubuntu 24.04 LTS.
- Windows 11 with WSL2 enabled
- Ubuntu 24.04 LTS installed on WSL2
- Coral USB Accelerator
- Admin access to the Windows machine
See https://github.com/ctch3ng/Installing-Different-Versions-of-Python-and-Managing-Virtual-Environments for details.
Install Required Python Packages (See https://coral.ai/docs/accelerator/get-started/#requirements for the latest instructions)
[WSL] Update 'apt-get' with Coral's REPO
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update
[WSL] Install the Edge TPU runtime.
sudo apt-get install libedgetpu1-std
Plug-in your Coral USB Accelerator. Reboot your WSL by hitting Windows Key + R to open Windows' Run command and run the following
wsl.exe --shutdown
[WSL] Install the pycoral library using the wheels, along with Pillow, and downgrade numpy to a compatible version:
python3.9 -m pip install https://github.com/google-coral/pycoral/releases/download/v2.0.0/tflite_runtime-2.5.0.post1-cp39-cp39-linux_x86_64.whl
python3.9 -m pip install https://github.com/google-coral/pycoral/releases/download/v2.0.0/pycoral-2.0.0-cp39-cp39-linux_x86_64.whl
python3.9 -m pip install Pillow
python3.9 -m pip install "numpy<2"
Note: The above code is for Python3.9. You can find other versions here https://coral.ai/software/#debian-packages .
Go to the usbipd-win release page https://github.com/dorssel/usbipd-win/releases/tag/v4.3.0 . Download the .msi installer file and run it to install.
Note: For details, visit https://learn.microsoft.com/en-us/windows/wsl/connect-usb .
[WSL] Open a WSL terminal and run:
sudo apt-get update
sudo apt-get install usbutils
[Windows] In Windows, open PowerShell as Administrator and run:
usbipd list
Note the BUSID of the Coral USB Accelerator (shown as Unknown). The following codes assume the BUSID is 1-4
[Windows] In the Windows PowerShell, run the following command to bind and attach the device to the default WSL VM:
usbipd bind --force -b 1-4
usbipd attach --wsl --busid 1-4
Note: If your default WSL is something else, you will need to run wsl --setdefault Ubuntu-24.04 to update it first.
[WSL] In WSL, run lsusb to confirm the device is listed. Look for “Google Inc.” in the list. If showing Global Unichip Corp., see Troubleshooting.
[WSL] Now, you can return to https://coral.ai/docs/accelerator/get-started/#3-run-a-model-on-the-edge-tpu to follow the procedures, which are replicated as follows.
mkdir coral && cd coral
git clone https://github.com/google-coral/pycoral.git
cd pycoral
bash examples/install_requirements.sh classify_image.py
python3.9 examples/classify_image.py \
--model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
--labels test_data/inat_bird_labels.txt \
--input test_data/parrot.jpg
[Windows] In PowerShell, run the following all at once while WSL is open and venv is activated. The line usbipd unbind -a needs to be executed twice on some systems to make it work.
usbipd unbind -a
usbipd unbind -a
usbipd bind --force -b 1-4
usbipd attach --wsl -b 1-4
usbipd list
[WSL] Run an example (it will fail, don't worry) [It appears that running this will trigger the device to change its ID]
Note: assume you are at ~\coral\pycoral\, and you have the examples installed. If not, go back to Run a model on the Edge TPU.
python3.9 examples/classify_image.py \
--model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
--labels test_data/inat_bird_labels.txt \
--input test_data/parrot.jpg
[Windows] Back to PowerShell, run the following all at once (again) while WSL is open and venv is activated.
usbipd unbind -a
usbipd unbind -a
usbipd bind --force -b 1-4
usbipd attach --wsl -b 1-4
usbipd list
[WSL] Run the example (again). This time, it should work. If not, repeat the process.
Note: assume you are at ~\coral\pycoral\.
python3.9 examples/classify_image.py \
--model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
--labels test_data/inat_bird_labels.txt \
--input test_data/parrot.jpg