Hands-on with Coral USB Accelerator (for Raspberry Pi)

Tags: google, coral, tpu, raspberry pi

Written on March 19, 2019

Coral launched dev board and usb accelerator a couple of weeks ago and they are now available on the store.

I though the usb accelerator would be enough for me, so I ordered one of it and tested on my Raspberry Pi 3B+.

---

1. Purchase

Fortunately, the shipping and delivery fee to my country was zero on Mouser.

Ordered on Saturday and received on Monday.

What an amazing delivery!

2. Unboxing

3. Configuration

Following this guide, I downloaded and unzipped edgetpu_api.tar.gz

and executed python-tflite-source/install.sh.

It installed several packages, changed udev rules, and copied library files.

The guide advices to plug in & out the usb accelerator, but to make sure, I just rebooted the machine.

4. Test

In the edgetpu directory, ran following command:

python3 demo/classify_image.py \
--model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
--label test_data/inat_bird_labels.txt \
--image test_data/parrot.jpg

and got almost the same result:

W0320 09:13:03.732145    6245 package_registry.cc:65] Minimum runtime version required by package (5) is lower than expected (10).
---------------------------
Ara macao (Scarlet Macaw)
Score :  0.61328125
---------------------------
Platycercus elegans (Crimson Rosella)
Score :  0.15234375

There was a warning message: Minimum runtime version required by package (5) is lower than expected (10)..

It doesn’t seem like a critical issue, but needs to be fixed sooner or later.

5. What now?

So, I got the usb accelerator and made sure it runs ok on my Raspberry Pi, but what to do now?

Maybe I need to learn TensorFlow Lite and related stuffs, train my own models, and build some applications with them.

Written on March 19, 2019