Tensorflow on hassio

Installation Home Assistant OS
43

You are spot on. I built the Tensorflow from the HASSIO add-on that Hunter had built, but it looks like that only supports AMD64. I’m using an Intel Core i7 chip.

https://github.com/hunterjm/hassio-addons/tree/master/tensorflow

1 Like
44

Intel i7 chips should work. It is called AMD64 because AMD created the orignial specification.

Source:

45

Thanks huge brain fart wow!

46

I built it from the HASSIO add-on that @hunterjm built.

47

You are probably not exposing the full CPU architecture through the VM. Found this Virtualbox issue ticket.
https://www.virtualbox.org/ticket/15471

It should work in Proxmox if you make the switch.

1 Like
48

Ran the install add-on from @hunterjm with no issues, but a few of us amd hassio users are getting the following error upon HA restart -

a
a1620×271 12.1 KB

With image_processing: removed, HA starts up just fine.

3 Likes
49

I’m getting this one too. Does anyone have any more information about it?

50

After upgrade to 0.94 tensorflow installed with jm addon stopped working. Error message

“AttributeError: module ‘tensorflow’ has no attribute ‘Graph’”

Working flawlessly until update.

51

So, had to uninstall the addon, reinstall, start addon. Restart HA. Then it worked!

Thanks to home assistant discord.

52

Hi @hunterjm, is still ok to follow the steps on this Dockerfile to build a custom tensorflow image ?

I just want to rebuild a custom image native to my cpu (without avx2 extension) for my Hassio VM instance.

Regards

53

i’m running hassio in docker on ubuntu 18.04.2 LTS x86_64, I installed the addon successfully however i still get this error after restarting:

Unable to locate tensorflow models or label map
not sure what i’m doing wrong

my yaml is as follows:

image
image1799×590 21.6 KB

54

@sunglass_guy Most likely because HA is not able to load the tensorflow library.
Check the line 93 at image_processing.py

There is a try / except that you have to remove to see in the logs that exacly is going on.

Just create a custom tensorflow component copying the tensorflow component files from HA GitHub repository to …/config/custom_component/tensorflow.

Then modify the file image_processing.py and replace the block from line 89 to 101 with this:

    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
    import tensorflow as tf # noqa
    from object_detection.utils import label_map_util # noqa

Reboot HA and check python errors on the HA logs.

55

thanks mate, looks like its working for me now on my i5 NUC

I do get the error: No OpenCV library found. TensorFlow will process image with PIL at reduced resolution

but I think this is a problem with the installer addon

56

thanks for the suggestion, but i didn’t get any new errors just the same as before:

2019-06-14 08:41:48 WARNING (MainThread) [homeassistant.loader] You are using a custom integration for tensorflow which has not been tested by Home Assistant. This component might cause stability problems, be sure to disable it if you do experience issues with Home Assistant.
2019-06-14 08:41:48 INFO (MainThread) [homeassistant.setup] Setting up image_processing
2019-06-14 08:41:48 INFO (MainThread) [homeassistant.components.image_processing] Setting up image_processing.tensorflow
2019-06-14 08:41:48 ERROR (SyncWorker_6) [custom_components.tensorflow.image_processing] Unable to locate tensorflow models or label map

I think its something on my end with the tensorflow files in /config/tensorflow/ that i grabbed from research, if I try to use those files hassio hangs during restart and never finishes is it possible they aren’t the right files for my hardware?

1 Like
57

@CountParadox I have my hassio runing on a ESXi VM instance and I was also able to finally install tensorflow building from sources. Now I am trying to figure out how to install that opencv library.

58

let me know how you go

59

Here are the steps I followed to get tensorflow worning on Hassio 0.94.3

In my case I have HA running on a VM so I decide to setup a separate VM with ubuntu+docker+portainer

Then on that aditionar VM, using portainer just pull the container homeassistant/amd64-base-python:3.7, get a bash session and run theese commands on it:

Once you get compiled the binary tensorflow-1.13.0rc2-cp37-cp37m-linux_x86_64.whl copy it to somewhere else, extract it and copy the tensorflow folder to Hassio at /config/deps/lib/python3.7/site-packages/tensorflow and restart HA

Update 1: I have created an alternative set of instructions
Update 2: For anyone who wants to safe the compile process, here you have a 7z file with all you need. Just extract the config content to you Hassio /config folder

60

I’ve got the same issue with this error showing up in the log after installing tensorflow on hassio running on ubuntu 18.04. Any luck getting this to work?

2019-06-19 16:12:42 ERROR (SyncWorker_3) [homeassistant.components.tensorflow.image_processing] Unable to locate tensorflow models or label map

61

Has anyone been able to get this running on a NUC w/ hassio? I’m still plagued by kernel-level trap code errors. Haven’t been able to find anything related to the actual cause.

62

After following these instructions, hass fails to boot. Any ideas?
Here’s the error message:

2019-06-22 20:55:34 WARNING (MainThread) [homeassistant.helpers.config_validation] Your configuration contains extra keys that the platform does not support.
Please remove [categories].  (See /config/configuration.yaml, line 300). 
Offending data: {"platform": "tensorflow", "source": "<redacted>", "model": "<redacted>", "categories": [{"category": "person", "area": {"left": 0.1}}]}