Would like to use the Google Coral USB dongle as an image processor ala the TensorFlow component. The coral can offload the inference to the usb device.
the edgetpu has a python api. https://coral.withgoogle.com/docs/reference/edgetpu.detection.engine/
Check out @blakeblackshear project he has been working on.
This uses a docker container running tensorflow instead of the HA component.
He has done a lot of work and it sounds like it might be what you are looking for - at least until HA has native support for something like Coral.
I have for sure, and it’s a slightly different use case. def following it as well.
I was using this before he started down the Coral path.
I ended up going with the built in HA component for Tensorflow instead.
This seemed to work better for what I was looking to do as I can control its activation.
I now use a motion detection and only trigger the Tensorflow object detection when motion is present.
Still have to work to configure it better to get less false triggers, but my SSD failed… so on hiatus until I replace it 
.
exactly, I just want to call it when needed, and was hoping to use the coral for that. still looking at options like maybe calling it via a shell command or something.
I haven’t researched the coral much - but it is basically to offload the CPU on graphic processing correct?
For my setup I simply used an FFMpeg binary motion sensor to trigger an automation that called the Tensorflow component. This worked well, except that the FFMpeg binary motion sensor decided to start going offline after a few days - as a temporary fix I added an automation to restart it if it went unavailable.
I use a Dahua camera, and just recently found out that the Amcrest component should work with Dahua’s. I plan to try this out instead of FFMpeg when I get my system back online.
Either way, I am assuming you mean to use the Coral for the processing (from my understanding of it) as I don’t believe the Coral would do anything with triggering - but I’ll admin I haven’t researched it.
the coral would do the inferencing offloading it from the main cpu. so it would replace the tensorflow component in your example workflow.
yes thanks!!!