Adding a custom model - day and night license plate recognition

[inspired27]

I am using docker version of Frigate and want to incorporate this excellent day and night license plate model by Mike Lud into Frigate can anyone help how to include to accompany the inbuilt models. For example which directory etc.

https://github.com/MikeLud/CodeProject.AI-Custom-IPcam-Models/tree/main/PT%20Models

[NickM-27]

Currently only one model is supported, you can't run multiple at the same time

[NickM-27]

I don't believe so, but of course the current Frigate model is just a Google demo model so you can download it and try to merge yourself.

The Frigate+ model already includes license plate as a label (though it's not out yet)

[inspired27]

When do you anticipate it being released? I would be happy to donate to accelerate the license plate recognition. Mike Lud dataset might help. Crime is happening regularly in our area with car theft, never used to be but is now and I need license plate recognition.

[NickM-27]

To be clear the model does license plate detection, ALPR is separate. I don't work on Frigate+ but I believe it was said on the HA discord taht the models will be generally available alongside 0.13

[inspired27]

Oh good that it at least identifies plates that's what I want Day and Night plate object detection then I will get home assistant to process the image thereafter. Is it available in the 0.13 dev version you put out for Cody's issue if so I will use the dev version and pay for Frigate+

[NickM-27]

No, it is separate.

[inspired27]

Thanks Nic, with Frigate+ when signed up is it all stored local in the docker container or does Frigate call out to the Frigate+ repository everytime to recognise objects? If local I am super keen to try it out are there any instructions to get up and running as license plate recognition more so at night is huge part of my use case.

…

________________________________ From: Nicolas Mowen ***@***.***> Sent: Saturday, 17 June 2023, 9:25 am To: blakeblackshear/frigate ***@***.***> Cc: crazyfrog-github ***@***.***>; Author ***@***.***> Subject: Re: [blakeblackshear/frigate] Adding a custom model - day and night license plate recognition (Discussion #6822) I don't believe so, but of course the current Frigate model is just a Google demo model so you can download it and try to merge yourself. The Frigate+ model already includes license plate as a label — Reply to this email directly, view it on GitHub<#6822 (reply in thread)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AICL2BLWDXU5KO2BR33PTPLXLTTOHANCNFSM6AAAAAAZIVUL34>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[NickM-27]

It runs locally just like any other model

[inspired27]

Okay so I finally figured it out, it is possible to add custom model via the deepstack detector plugin in. I am running dev 0.13.0-0996883 with deepstack plugin pointing to codeproject.ai server which is running okay. However the license-plate model in the codeproject.ai server results in return of Dayplate or Nightplate with capitalise first letter and is not being passed back to Frigate. Please can someone confirm what I need to add to the Frigate config file (and if capitalised or no caps or both), and do I need to update labelmap.txt if so with what?

detectors:
  deepstack:
    api_url: http://192.168.1.xx:32168/v1/vision/custom/license-plate
    type: deepstack
    api_timeout: 1 # seconds

[kirsch33]

Okay so I finally figured it out, it is possible to add custom model via the deepstack detector plugin in. I am running dev 0.13.0-0996883 with deepstack plugin pointing to codeproject.ai server which is running okay. However the license-plate model in the codeproject.ai server results in return of Dayplate or Nightplate with capitalise first letter and is not being passed back to Frigate. Please can someone confirm what I need to add to the Frigate config file (and if capitalised or no caps or both), and do I need to update labelmap.txt if so with what?

detectors:
  deepstack:
    api_url: http://192.168.1.xx:32168/v1/vision/custom/license-plate
    type: deepstack
    api_timeout: 1 # seconds

see this comment:

#6143 (comment)