Image processing with USB acceleration - all pi! ARCHIVED

jasondreher · January 28, 2020, 2:41pm

OK, I’ll swap out the power supply and see if that helps, if not I will create an issue. Thanks!!

renat · February 3, 2020, 9:53am

Hi. Can be used with Coral M2 accelerator?

robmarkcole · February 3, 2020, 1:33pm

I think someone on this thread might have tried that. In principle there should be no problem

Dreamoffice · April 29, 2020, 5:56am

Hi
Is the hassio add on released already?
I got my coral stick yesterday. Now I want to get started. I’m running hassio on a raspberry 4. Which way should I go?

robmarkcole · April 29, 2020, 6:25am

There is no Hassio add on currently, that requires some work on Hassio to support the hardware. Perhaps @frenck can provide some guidance on what needs to be done to support the coral stick?

Dreamoffice · April 30, 2020, 10:12am

how would you set it up? on a separate raspberry? or can i use the raspberry where hassio is running on?

robmarkcole · May 1, 2020, 11:40am

Hassio occupies the whole pi, and getting hardware to work with it is a significant challenge, I understand (but have not tried!)

Dreamoffice · May 1, 2020, 1:11pm

i set it up and it is running with the default detector. as soon as i changed the detector the service will stop and i got a message that the process uses more than 10% of system resources.

for me the default detector is fine at the moment.

does anybody has setup a notification service which sends the picture with html5_push to the connected devices?

Auto101 · May 11, 2020, 10:47pm

Hi
I finished the installation but I get this error when I try to start the server.

Any suggestions?

pi@raspberrypi:~/coral-pi-rest-server $ python3 coral-app.py
Traceback (most recent call last):
  File "coral-app.py", line 103, in <module>
    engine = DetectionEngine(model_file)
  File "/usr/local/lib/python3.7/dist-packages/edgetpu/detection/engine.py", line 72, in __init__
    super().__init__(model_path)
  File "/usr/local/lib/python3.7/dist-packages/edgetpu/basic/basic_engine.py", line 40, in __init__
    self._engine = BasicEnginePythonWrapper.CreateFromFile(model_path)
RuntimeError: Could not open '~/Documents/GitHub/edgetpu/test_data/mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite

stalr5 · May 11, 2020, 11:06pm

I got the same error, changed the path in coral-app.py to:
“–models_directory”,
default=“/home/pi/all_models/”,
and it worked, not sure if this is the right thing to do, i,m still setting things up

Auto101 · May 12, 2020, 1:14am

Thanks , that fixed the issue .

I have few questions on the next steps integrating this with HA . I have 2 Pi’s one for the coral-pi-rest server and the other for HA instance . Now, I am reading HASS-Deepstack-object GitHub page and looks like this needs to be run a different server ?? Is that true in my case?

Excuse me for my questions as I am new to this

robmarkcole · May 12, 2020, 7:59am

The coral pi API mirrors the deepstack API, so they are swappable

Auto101 · May 12, 2020, 7:13pm

Thanks Robin!

What would be the next steps for me to integrate this to HA without the deepstack as I am using the coral?

Once I integrate the coral with HA , I plan to do an automation for a switch using node-red .
The automation would be something like to turn on a switch when the camera sees a person.
Appreciate if you could forward any documentation which helps in implementing this .

robmarkcole · May 13, 2020, 6:58am

I am making some improvements to deepstack integration, then will pick up the coral pi again

Auto101 · May 13, 2020, 11:25am

Great, thanks for the update !

robmarkcole · May 15, 2020, 9:28am

some people are hoping the coral code i wrote will perform as well as the official deepstack docker container - that is not the case. The coral code is very alpha, and doesn’t handle multiple requests

jazzmonger · May 15, 2020, 4:17pm

@robmarkcole
Yeah, that may be the case, but it still works great!

Auto101 · May 19, 2020, 2:46am

Agreed , it works great !

Has anyone tried training a custom model from google teachable - https://teachablemachine.withgoogle.com/
?

I exported the edgetpu tflite for a custom model . When I try to run the same model with this script , I get below error :
ValueError: Detection model should have 4 output tensors!This model has 1

robmarkcole · May 19, 2020, 3:11am

I am working on support for custom models now…!

jazzmonger · May 20, 2020, 8:16pm

@robmarkcole

re: Coral rest server

there any way to make the font larger and/or a different color background on the boxes? I’m not able to read anything on any detections, possibly b/c I’ve reduced the camera image to VGA resolution (and the bitrate is now 96kbps on the sub stream) as its a wifi camera and the link isn’t great. This solved some of my delay problems and timeouts:

I tried modifying this bit of code and restarted, but it doesn’t have any effect:

I’d like to move them to the bottom instead of the top of the image as well.

If you can just point me to the module that handles these drawing routines, I should be able to figure it out from there.

Jeff