IP Cam Talk Custom Community DeepStack Model

@shannondalebreaux You can find @MikeLud1 models on his github repo: MikeLud - Overview

If you use CodeProject.AI (with BI5), Mike's models are included so no separate download is necessary.
I'm using deep stack. CP AI kept crashing and timing out on me. DS never did that to me. Not sure why.. Thanks I'll try to download add and use them with deep stack...
 
  • Like
Reactions: gwminor48

Attachments

  • Screenshot_20230606-090402_TeamViewer.jpg
    Screenshot_20230606-090402_TeamViewer.jpg
    620.2 KB · Views: 75
Have been using DS with these models for the last year or so, @MikeLud1 would you suggest changing from the "combined" model to the general?

Also, in general, haven't yet tried CodeProjectAI. Is it at a point where you feel it does a better job than DS for ALPR and general AI (person/car detection)? Haven't really checked out to use it vs DS.

1686499550969.png
 
I just started using these custom models and stumbled on a few things..

First I assume I want to add objects:0 to my custom model list, or uncheck use default objects on the main AI setting page..

Second, am I correct in not needing ipcam-general and ipcam-combined at the same time? When I first started using these models I had both and was getting a ton of false positives. I've since removed ipcam-general and use only combined, and delivery on certain cameras.
 
Hi all, just a question about BI+CPAI, just curious...

After disabling AI for a cam in Blue Iris in camera settings (alert confirmation=none in Artifical Intelligence setup page), I see Blue Iris still using AI for that camera, analysing only one image per trigger. I've confirmed it looking at CPAI logs. Why?
 
Hey guys, I know a lot of us are switching to CPAI and I plan on trying it, but while researching it, I came across some questions regarding my Deepstack set up. It's been a while since I messed with it and have some questions regarding the models.

What are the correct contents of the general and dark models?

This link - https://ipcamtalk.com/threads/ip-cam-talk-custom-community-deepstack-model.60376/ - shows vehicle and not car and truck in the general model
This link - https://user-images.githubusercontent.com/36526272/171761824-1d3252ee-0273-49b4-b74f-c9554f68e55f.png - shows car and truck and not vehicle in the general model

So which is most current? Currently I'm using the general, animal and dark models. I have the To Confirm field set up like this - person,vehicle,bird,cat,dog,bear,deer,rabbit,raccoon,fox,skunk,squirrel,Car,Cat,Dog,People. I'm also not sure where I got People from as both links show the Dark Label containing Person and not People.

This leads me to the Mark as vehicle field. Do I put car,truck,Car or vehicle,Car

Lastly, can someone explain the attached picture to me? This is just a manual trigger I ran without anything in the FOV. I just wanted to see the times it took to analyze the picture. Starting at the top of the picture, I understand the entries that say, general (61 msec) dark (93 msec) and animal (48 msec). This is the time it took for those models to analyze the picture. What does the T+2536 msec mean? I know the [207 msec] to the right of the T+2536 is the total time it took all three models to run. I'm just not sure what the T+2536 means.


I appreciate you guys looking,
Chris
 

Attachments

  • Deepstack Analysis.JPG
    Deepstack Analysis.JPG
    37.4 KB · Views: 30
I totally forgot I started playing around with the delivery models. I have them installed and configured globally and per camera. But I'm never getting them identified. Under the AI tab on a alert, I'm seeing "Unable to create YOLO detector for model Delivery"

I am using the onnx files since I'm using .net for detection. And under the global settings, custom models does list delivery as being loaded.

[EDIT] Whatever that error is, I just ignored and played back a clip of the mail truck and turned on analyze with AI. It is working because I see the USPS boxes highlighting the logo. But in the final alert it only shows truck. Perhaps it's just not seeing it in the 3 images I have AI set to analyze
 
Last edited:
I totally forgot I started playing around with the delivery models. I have them installed and configured globally and per camera. But I'm never getting them identified. Under the AI tab on a alert, I'm seeing "Unable to create YOLO detector for model Delivery"

I am using the onnx files since I'm using .net for detection. And under the global settings, custom models does list delivery as being loaded.

But I get this in the server console, not sure why it's looking for the .pt file. IIRC .net doesn't use them, but the onnx models
15:42:32:Object Detection (YOLOv5 6.2): C:\Program Files\CodeProject\AI\modules\ObjectDetectionYolo\custom-models\delivery.pt does not exist
15:42:32:Object Detection (YOLOv5 6.2): Unable to create YOLO detector for model delivery

@MikeLud1 created the .onnx port of the delivery model. I've not tried using it so I'm not sure what to recommend.

Edit: I might try separating out the processing so you just do delivery on a clone. Also, I assume it works if you take an image and test it directly in CodeProject.AI
 
Last edited:
  • Like
Reactions: scoob8000
You can download the delivery model from the below link


I have it loaded, and it seems to be kinda working. Some of the images analyzed it sees a positive result. Some get that unable to create yolo detector error. But even the ones that work are higher than my 50% threshold but I still never see those in my alert list.

Maybe it's how I have the alert AI tab setup?
 

Attachments

  • remmina_75_10.40.40.75_20231027-121314.814187.png
    remmina_75_10.40.40.75_20231027-121314.814187.png
    749.7 KB · Views: 24
  • aisettings.png
    aisettings.png
    49.5 KB · Views: 24
Switching to the General model. I must not have these set up properly, as I'm not getting any alerts?

1698578245322.png