5.5.8 - June 13, 2022 - Code Project’s SenseAI Version 1 - See V2 here https://ipcamtalk.com/threads/codeproject-ai-version-2-0.68030/

I can't seem to get my custom models to work. If I check the box for "default object detection" it works but uses the standard model. Make sure you uninstall 1.5 and then update to 1.55 and restart everything. Mine then populated the custom folder path, which is : C:\Program Files\CodeProject\AI\AnalysisLayer\CustomDetection\assets
 

Attachments

  • Screenshot 2022-07-16 094046.png
    Screenshot 2022-07-16 094046.png
    84 KB · Views: 97
Last edited:
It finds the combined model to benchmark, but something still has changed ?
 

Attachments

  • Screenshot 2022-07-16 094656.png
    Screenshot 2022-07-16 094656.png
    70.9 KB · Views: 62
So I think I have it working now. I had to modify two lines in the windows registry to enable and define the custom model folder path.

Screen Shot 2022-07-16 at 9.48.09 AM.png

The custom models are already included in the install and are located C:\Program Files\CodeProject\AI\AnalysisLayer\CustomDetection\assets directory so there is no need to download them separately. However, the naming convention is different so you'll need to make sure to use the correct names for the model you want to use.

Screen Shot 2022-07-16 at 9.53.09 AM.png
Here's how i have my AI trigger configured for my LPR camera.

Screen Shot 2022-07-16 at 9.57.50 AM.png

And here are the results using the ipcam-general model.

Screen Shot 2022-07-16 at 10.01.43 AM.png

Global AI settings with default object detection disabled.

Screen Shot 2022-07-16 at 10.03.53 AM.png
 
Last edited:
So I think I have it working now. I had to modify two lines in the windows registry to enable and define the custom model folder path.

View attachment 133658

The custom models are already included in the install and are located C:\Program Files\CodeProject\AI\AnalysisLayer\CustomDetection\assets directory so there is no need to download them separately. However, the naming convention is different so you'll need to make sure to use the correct names for the model you want to use.

View attachment 133663
Here's how i have my AI trigger configured for my LPR camera.

View attachment 133667

And here are the results using the ipcam-general model.

View attachment 133669

Global AI settings with default object detection disabled.

View attachment 133671

Interesting, My registry was correct and I never had to mess with it, but using the ipcam-combined did make the model work...Thanks !
 
You need to use licence-plate.
Thanks!

Edit: I got it working, but it's not very reliable. What would happen if I replaced the .pt file with yours and changed the confirmed section to DayPlate? Will it still work?
 
Last edited:
  • Like
Reactions: gwminor48
Now it's time to add the USPS model so that I can have my home automation send mail delivery notifications.
The USPS Model also works well with SenseAI.

Screen Shot 2022-07-18 at 9.05.40 PM.png

Even with the additional models the CPU processing is significantly lower than with DeepStack. The graph below is the CPU load of my BI computer running 16 cameras (+1 cloned) of which 14 have SenseAI enabled.

Screen Shot 2022-07-18 at 9.12.37 PM.png

With DeepStack my CPU load averaging well above 75%.
 
The USPS Model also works well with SenseAI.

View attachment 133897

Even with the additional models the CPU processing is significantly lower than with DeepStack. The graph below is the CPU load of my BI computer running 16 cameras (+1 cloned) of which 14 have SenseAI enabled.

View attachment 133898

With DeepStack my CPU load averaging well above 75%.
What kind of processing times are you seeing on any given camera/event (given that you are running CPU only, and with custom models)? @MikeLud1 mentioned that in his testing, the times seemed oddly long on this current (1.5.5) version compared to one generation back, and that he had raised the issue with the developers. Just looking to see your times for comparison.
 
What kind of processing times are you seeing on any given camera/event (given that you are running CPU only, and with custom models)? @MikeLud1 mentioned that in his testing, the times seemed oddly long on this current (1.5.5) version compared to one generation back, and that he had raised the issue with the developers. Just looking to see your times for comparison.
Most models are processing between 500 and 550 msec which is fast enough for what I'm using them for.
 
  • Like
Reactions: jrbeddow
You need to edit the .json file . Here is what you edit to disable Face detection for example. Just change it from true to false. Careful not to get trigger happy. You can always save a backup copy first.
 

Attachments

  • Screenshot 2022-07-27 165041.png
    Screenshot 2022-07-27 165041.png
    48.5 KB · Views: 90
  • Screenshot 2022-07-27 165133.png
    Screenshot 2022-07-27 165133.png
    69.9 KB · Views: 90