Face and person detection with Deepstack - local and free!

Share your Projects!
383

Its already on HACS

384

Apologies, my mistake/

1 Like
385

Oh it needs to be added as a custom install?

How, havenā€™t tried that yet?

386

In HACS settings add a custom repository with the Github path for the project then it will appear as installable.

387

@juan11perez I updated deepstack face to v0.2 which allows use of an API key to secure deepstack. There is also an optional fast detect_only mode.
Cheers

388

@robmarkcole thank you very much. Updated with no issues.

389

@juan11perez any feedback let me know, are you using it instead of Facebox?

390

@robmarkcole right now Im running both with similar results

1 Like
391

Hi,

Super keen to give this a go, but donā€™t seem to be able to locate the noavx version of the Docker image. No doubt Iā€™m missing something really simple.

Iā€™m planning to run it on a Qnap Nas device.

Thanks for any input.

.N

392

If you use the correct docker run command (in the deepstack docs) it will pull the image

393

Hi Robin,

Thanks for your reply.
Qnap has an app (Container Station) which is a GUI that integrates both LXC and Docker virtualization technologies. It provides the ability to search for an image, but also a dialog to Pull a particular image. Unfortunately the ā€œImage Nameā€ field doesnā€™t allow you to use a colon as in deepstack:noavx
I will have to try doing it from the cmd line.

Nigel

1 Like
394

Is anyone using multiple targets? Tried this but it didnā€™t work.

image_processing:
  - platform: deepstack_object
    ip_address: localhost
    port: 5000
    scan_interval: 20000
    save_file_folder: /config/www/deepstack_person_images
    target: person, truck
    confidence: 80
    source:
      - entity_id: camera.front
        name: person_detector
395

I tried the same, but I ended up doing two instances of the image_processing.

I donā€™t think the other one is possible at the moment.
But the other way around works perfectly!

396

Not at home atm but I ended up doing this by editing the code very slightly.

In the initial config I still had target:person but in the code did a simple edit in the section which checked for prediction = target to add in AND prediction = ā€˜carā€™. This was more a bandaid fix but enough for my purposes. Now either person or car will trigger the save file event and boxes drawn on both.

For me this helped identify when a car or person (or both) are in the image.

A better solution would be to pass in target as an array and go from there but this would need more code editing.

397

There is only a single target per entity but you can configure multiple entities

398

Does this mean that the same image would need to be processed twice ie once to detect ā€˜personā€™ and once for ā€˜truckā€™.

399

Yes each entity will process the same image in this case. Alternatively you can use the image_processing.object_detected event

400

Hi
Iā€™m running the latest releaseHass.io under Ubuntu in a Docker environment. After setting up Deepstack based on the guide Iā€™ve tried to restart HA but it gave me this error:

Platform error image_processing.deepstack_face - No module named ā€˜deepstackā€™

Could you please help me correcting this error? Iā€™m clueless.

401

hi rob, great work again. in your opinion is this better for person detection than tensorflow? I ran out of free images (5000) with amazon rekognition and moved to local tensorflow recently (NUC with i5 8th gen). please let me know. Thanks.

402

Well I wrote Deepstack integration for a reason :slight_smile: In terms of performance there is not much difference. Deepstack easier to use IMO

1 Like