Getting Started with the Google Coral Dev Board Micro - Low-Power Object Detection | DigiKey
Ғылым және технология
Dive into the world of AI with our latest tutorial featuring the Google Coral Dev Board Micro! This microcontroller board packs a punch with its dual-core ARM Cortex-M7 and M4, complemented by a Tensor Processing Unit (TPU) AI accelerator. We guide you through the process of training a custom object detection model (MobileNetV2-SSD) and deploying it to the Coral Micro, achieving remarkable inference speeds of 8-12 frames per second using 320x320 color input images, all for under 1 watt.
A written step-by-step guide for this tutorial can be found here: www.digikey.com/en/maker/proj...
Get the Coral Dev Board Micro here: www.digikey.com/en/products/d...
Code for this tutorial can be found here: github.com/ShawnHymel/google-...
We start by collecting data from the Coral Micro board and labeling the data using www.makesense.ai/. We then train our custom object detection model using MediaPipe. MediaPipe is Google’s workflow for training common machine learning models using custom data and deploying them to a variety of systems. At the end of the training process, you will produce a variety of model formats for TensorFlow. While you can deploy them to any number of systems, we specifically need the TensorFlow Lite (TFLite) model compiled and optimized for a TPU.
From there, we deploy the TFLite model to the Coral Micro with a custom app written using the FreeRTOS-based coralmicro SDK. The firmware takes images from the onboard camera and performs inference using the optimized model running on the TPU. Bounding box information is then streamed over the serial port (USB and UART pins) as well as out through an optional web server that can be accessed via USB. By browsing to the Coral Micro’s server, you can visualize what the camera sees along with bounding box information to help you debug your model.
Whether you're testing, developing, or debugging, this tutorial provides all the tools and knowledge you need to make the most out of the Google Coral Dev Board Micro!
MediaPipe documentation: developers.google.com/mediapi...
Example of training an object detection model using MediaPipe: developers.google.com/mediapi...
Getting started with the Coral Dev Board Micro: coral.ai/docs/dev-board-micro...
coralmicro API documentation: coral.ai/docs/reference/micro/
Learn more:
Maker.io - www.digikey.com/en/maker
DigiKey’s Blog - TheCircuit www.digikey.com/en/blog
Connect with DigiKey on Facebook / digikey.electronics
And follow us on X / digikey
0:00 Intro
0:57 Flash image collection code
6:06 Collect images
7:00 Label bounding boxes in images
10:41 Train object detection model
20:05 Test object detection model
25:23 Deploy model to Google Coral Micro
29:17 Test live object detection inference
Пікірлер: 14
Get the Coral Dev Board Micro here: www.digikey.com/en/products/detail/okdo/G650-07968-01/22121650
@josephchamness9915
4 ай бұрын
This link or the one above don't work for me.
@ShawnHymel
4 ай бұрын
@@josephchamness9915 Are you getting an error message or code (e.g. 403 or 404)? I just checked, and it's working for me.
Awesome, always love this guy's videos
I have a feeling this is a lot like the Windows 95 stage of Micro AI, looking forward to the day which will soon arrive when this is as easy as making a PIR sketch on an UNO
It would be nice to compare Google Coral with Alif chip which has two Cortex M55 with Ethos U55 ML accelerator on the board.
Can this board be also used a USB accelerator with Raspberry Pi?
Interesting hardware, but it is surprising to see such poor video quality, this alone makes it practically unusable for a lot of computer vision applications.
Description When i run lsusb Google Inc. Coral Dev Board Micro I already tried everything on the flash tool troubleshooting section as start the serial mode. When i run python3 flashtool.py -e detect_faces, the error appears Finding all necessary files Creating Filesystem Section: 0x0 iMX bootable image generated successfully Boot Section 0x00000000: FILL | adr=0x00010000 | len=0x00000004 | ptn=0xc2000105 FILL | adr=0x00010004 | len=0x00000004 | ptn=0x00010020 FILL | adr=0x00010008 | len=0x00000004 | ptn=0x00040004 FILL | adr=0x0001000c | len=0x00000004 | ptn=0x00080004 FILL | adr=0x00010020 | len=0x00000004 | ptn=0xc0010021 ENA | adr=0x00010000 | cnt=0x00000004 | flg=0x0110 ERAS | adr=0x00000004 | cnt=0x00000004 | flg=0x0110 ERAS | adr=0x00000008 | cnt=0x00000004 | flg=0x0110 ERAS | adr=0x0000000c | cnt=0x00000040 | flg=0x0110 LOAD | adr=0x00000004 | len=0x00017854 | crc=0x511a1083 | flg=0x0110 LOAD | adr=0x00000008 | len=0x00017854 | crc=0x85117454 | flg=0x0110 STATE_CHECK_FOR_ANY STATE_CHECK_FOR_ELFLOADER STATE_RESET_ELFLOADER Traceback (most recent call last): File "/home/augusto/.coral/coralmicro/scripts/flashtool.py", line 1193, in main() File "/home/augusto/.coral/coralmicro/scripts/flashtool.py", line 1182, in main RunFlashtool(StateCheckForAny, sbfile_path=sbfile_path, File "/home/augusto/.coral/coralmicro/scripts/flashtool.py", line 821, in RunFlashtool (prev_state, state) = (state, state(**state_kwargs)) ^^^^^^^^^^^^^^^^^^^^^ File "/home/augusto/.coral/coralmicro/scripts/flashtool.py", line 610, in StateResetElfloader with OpenHidDevice(ELFLOADER_VID, ELFLOADER_PID, serial_number) as h: File "/usr/lib/python3.12/contextlib.py", line 137, in _enter_ return next(self.gen) ^^^^^^^^^^^^^^ File "/home/augusto/.coral/coralmicro/scripts/flashtool.py", line 589, in OpenHidDevice h = hid.device() ^^^^^^^^^^ AttributeError: module 'hid' has no attribute 'device'. Did you mean: 'Device'?
This board is expensive. $104 per box. If I want to experiment with object detection I would buy an orange pi 5 that have tensa core too for few dollars more.
Cool info, but after their AI debacle, I have no interest in spending money with them or using their products.
@glennimmanuel9338
3 ай бұрын
why ?
@frankhandley7648
3 ай бұрын
@glennimmanuel9338 their inner racism was so strong it made a joke of their AI.