Tensorflow on hassio

Installation Home Assistant OS
63

Yes, I can confirm that this is running on my NUC, although my HASSIO is installed on Ubuntu 18.04 LTS VM on ESXi 6.7.

64

Thanks for the confirmation. I have a headless Debian 9 VM on Proxmox running a homeassistant/intel-nuc-homeassistant docker image, but even after compiling a binary according to the instructions from @JBelinchon (against homeassistant/amd64-base-python:3.7), Iā€™m still getting vague trap core errors, core dumps, and the HA container doesnā€™t start. Canā€™t find anything either in the containerā€™s console or HA logs that indicates what aspect of tensor is ending the execution.

65

Did you manage to solve this? I have the same issueā€¦

66

No luck here

67

I ended up switching over to the custom AWS Rekognition component.

68

if that helps anyone, to resolve the [homeassistant.components.tensorflow.image_processing] Unable to locate tensorflow models or label map message, I ended up doing this:

  1. Follow instructions here
  2. Manually install tensorflow (pip3 install tensorflow)
  3. Run this bash script
  4. Add tensorflow to config as per here

No errors anymore and tensorflow image_processing sensors now appear in HA.
I now need to play about with it to see if it works so that I can hopefully say goodbye to unwanted notifications because stupid seagulls peck on my lawn :open_mouth:

69

I get the same error. Did you ever find a solution?

70

I get this too. Itā€™s a known ā€œissueā€. Actually not an issue as it still works as itā€™s supposed to do you can ignore this

71

you can manually install OpenCV but it will run without it

72

But it will produce better images with it, correct? Can you share instructions on how to install opencv on hassio? Thanks

73

I donā€™t think it will produce better images, it will just prices the camera feed better or something

Do you have access to the homeassistant container?

74

I can ssh into hassio

75

So it works with Intel and AMD chipsets?

76

Iā€™m having the exact same issues that @HITChris had. Iā€™m running on a NUC 8 i5, Proxmox VM using the install method from whiskerz007 as highlighted on Dr Zzs youtube. Iā€™m not sure if itā€™s something to do with exposing the right properties of my system to Proxmox, but any help would be appreciated. @HITChrisā€™s solution was to change hardware, and Iā€™m not up for that.

77

Probably just a wording thing, but I simply switched to a different component for the image_recognition platform (cloud Rekognition vs local Tensor). No hardware change, and besides the half second delay it takes for a request/response, itā€™s worked out great.

78

OK, my bad Chris. Thanks for the reply. If I donā€™t get anywhere here, Iā€™ll look into it.

79

Hello, I have installed in my NUC with the addon and the files, but when I put in the configuration.yaml the ā€œimage processing componentā€, my hassio does not start the UI.
If I remove the image processing component and all works correctly.
what am i doing wrong?

No addon or log errors

My configuration is NUC Ubuntu 18.04.3 + Docker + hassio

 - platform: tensorflow
   scan_interval: 20000
   source:
     - entity_id: camera.xiaofang_1
     - entity_id: camera.xiaofang_2
     - entity_id: camera.prueba
   file_out:
     - "/config/www/tensorflow/{{ camera_entity.split('.')[1] }}_latest.jpg"
     - "/config/www/tensorflow/{{ camera_entity.split('.')[1] }}_{{ now().strftime('%Y%m%d_%H%M%S') }}.jpg"
   model:
     graph: /config/tensorflow/faster_rcnn_inception_v2_coco_2018_01_28/frozen_inference_graph.pb
     categories:
       - person
       - dog
80

@rabeliyo, Iā€™m having the same issue. Did you install hassio on the NUC directly from the default install on the home-assistant website?

81

@thefarelkid, I have tried 2 types of installations, the special one for NUC from the official hassio website, and the one I currently have with ubuntu + docker + hassio, the latter, I followed the instructions on this site

Neither installation lets me start if I put image processing component

I have an Intel Celeron N3050 and 8Gb RAM, so I donā€™t think itā€™s a problem of requirements

82

Anything in the logs? The path to the model is correct?
This is what i have:


image_processing:
  - platform: tensorflow
    source:
      - entity_id: camera.back_yard
      - entity_id: camera.terrace
    confidence: 75
    scan_interval: 10000
    file_out:
      - "/config/www/tensorflow/{{ camera_entity.split('.')[1] }}_latest.jpg"
    model:
      graph: /config/tensorflow/coco/frozen_inference_graph.pb
      categories:
        - category: person