Face and person detection with Deepstack - local and free!

robmarkcole · May 5, 2020, 11:15am

Deepstack is now 100% free

sunny1283 · May 5, 2020, 11:29am

Wow that’s big news, how will they earn revenue though to keep enhancing and keeping it alive?

jonwilliams · May 5, 2020, 12:46pm

It seems to have stopped working since they release the free version. Noticed 401’s in the container logs so checked the docs and they have removed the API key env variable…I have updated my stack and removed the API key from both the compose file and the HA config…

Now see that its returning 200’s:

Container logs:
[GIN] 2020/05/05 - 12:37:48 | 200 | 194.66148ms | 172.16.13.4 | POST /v1/vision/detection

Here is my HA config:

image_processing:
  - platform: deepstack_object
    ip_address: deepstack
    port: 5000
    #api_key: secret! deepstack_api
    #scan_interval: 30 # Optional, in seconds
    confidence: 50
    save_file_folder: /config/www/front/
    save_timestamped_file: True
    source:
      - entity_id: camera.front
        name: deepstack_person_detector_front
  - platform: deepstack_object
    ip_address: deepstack
    port: 5000
   #confidence: 50
    #api_key: secret! deepstack_api
    #scan_interval: 30 # Optional, in seconds
    save_file_folder: /config/www/back/
    save_timestamped_file: True
    source:
      - entity_id: camera.rear
        name: deepstack_person_detector_back

Any ideas?

Ta in advance!

robmarkcole · May 5, 2020, 1:44pm

@sunny1283 the business model is to provide free use for individuals, and make money from enterprise customers with special needs.

@jonwilliams what is the problem, all looks fine?

jonwilliams · May 5, 2020, 2:04pm

Hi,

Ok - now I have removed the api key (and rebooted HA!); the one camera is working and the ‘last_person’ image is being saved. My other camera (back) isn’t! While I see the 200 in the Deepstack container logs, it’s not detecting ‘people’ for some reason.

Ta

Jon

jonwilliams · May 5, 2020, 2:58pm

It seems that it is actually working - however its hit and miss now.

robmarkcole · May 6, 2020, 3:31am

Please note, we are in config talking about the api_key which you can configure to secure the deepstack endpoint. This should not be confused with the ‘product key’ that was previously required to ‘activate’ your deepstack software.
Note I did just try out the noavx image and it doesn’t appear to be updated yet. I am personally waiting to see the open sourced code before making any judgements about how this product will develop.
Cheers

jonwilliams · May 6, 2020, 7:35am

Cheers, seems to be working now. But I have had to disable Face detect and recognise for some reason. Not too bothered about that anyway.

kriss_4 · May 7, 2020, 6:24am

Hi have hassio installation and follow the guide robmarkole guide.
Now whe i try to call the service service image_processing.scan with image_processing.deepstack_person_detector i receive this error:

"Logger: homeassistant.components.image_processing
Source: components/image_processing/init.py:128
Integration: Elaborazione immagini (documentation, issues)
First occurred: 8:18:49 (1 occurrences)
Last logged: 8:18:49

Error on receive image from entity: Camera not found"

The deepstack is activated and my configuration is:

camera 11:

platform: local_file
file_path: /config/snapshots/test.jpg

image_processing:

platform: deepstack_object
ip_address: 192.168.1.12
port: 5000
save_file_folder: /config/deepst
save_timestamped_file: True
target:
- person
- car
  source:
- entity_id: camera.local_foto
  name: deepstack_person_detector

Any help please.
Thank you

kriss_4 · May 7, 2020, 4:16pm

I’ve made some more test and the curl command picture is working as aspect.
There is a problem in the hassio side. the entity is unknow.
end the message is still : Error on receive image from entity: Camera not found.
If i’m implementing the card on hassio, i’m able to see the image.

kriss_4 · May 7, 2020, 5:06pm

OK. Found the issue my self again.
thx

monkeydust · May 8, 2020, 8:07am

Hi ArnoNel - did you ever figure out how to do this? Just watched https://www.youtube.com/watch?v=fwoonl5JKgo - awesome setup but I am on Ring and dont want to switch away just yet.

Dayve67 · May 11, 2020, 7:58am

Hello @robmarkcole

I see you have done a version update with some breaking changes. Is there anything I need to do with my configuration before I apply the update?

robmarkcole · May 11, 2020, 8:43am

Yes, as described in the release notes

Dayve67 · May 11, 2020, 1:08pm

Thanks. I’ll do the update and see if anything breaks.

robmarkcole · May 11, 2020, 1:50pm

Please keep discussion to deepstack topic

Yoinkz · May 12, 2020, 8:25am

Hi,

It seems that after updating the latest Deepstack compoment, it provides me with old pictures and not new ones. I tried looking in the documentation and the only thing I saw that was changes was the “target” to “targets”.

What else have I missed?

EDIT: I found the reason. So it looks like the file is now also being named xxxxxxxx_latest_person but now just latest.

robmarkcole · May 12, 2020, 3:55pm

Yes, since there can be multiple targets, the previous convention didnt make sense any more

jazzmonger · May 12, 2020, 5:51pm

I just upgraded to the new version 3.0 and I’m having all kinds of trouble. I changed target to target(s) and read the docs and realized you removed scan_interval: also, so I got rid of that. The object loads ok, and I can only trigger it VERY sporadically. Sometimes it fires, most of the time it doesn’t which now renders my entire integration unusable. it might take 20 clicks on “Call Service” to fire once. The camera works fine and the stream is stable and consistant.

Here is my config.

  - platform: deepstack_object
    ip_address: 192.168.1.208
    port: 5000
    api_key: !secret deepstack_api
#    scan_interval: 10
    confidence: 65
    save_timestamped_file: false
    save_file_folder: /config/www/snapshots
    targets:
      - car
      - truck
      - person
    source:
      - entity_id: camera.frontgate
        name: frontgate_object_detection

and clearly it is configured and works ok. Sometimes…

Also, you mention " use Developer tools -> EVENTS -> :Listen to events , to monitor these events". When I can get it the object to fire using the call service, there are no events listed at all, even though I can fire that event manually as often as I want:

I did find this in the logs:

Source: __main__.py:347 
First occurred: 10:13:47 AM (21 occurrences) 
Last logged: 10:47:48 AM

Update of image_processing.frontgate_object_detection is taking over 10 seconds

Any thoughts before I go back to v2.7? this is driving me nuts.

Jeff

RG3 · May 12, 2020, 6:46pm

Jeff,

I’ve updated to 3.0 and made the same changes as you have identified and haven’t had any issues with the image scan firing from the developer tools services page, an automation or node red. Comparing your config to mine the only difference I have from you is I don’t include the api_key line. I doubt that has any role in your issue, but your config looks good to me. Can you view the CPU usage of your deepstack instance and see if there is a corresponding CPU spike to you calling image processing scan service? I know that is one way I test that the everything is working together. The intermittent nature of your issue makes it all the harder to pin down.