CodeProject.AI Version 2.5

[Cooltiger]

I am on 5.8.8.2 and custom model list is working fine.

My BI on 5.8.8.2 also reverted to Auto and not delayed start.
My BI on 5.8.7.11 also had reset to Auto not delayed start.

 

[CCTVCam]

Yes, the difference in accuracy is due to using default model yolo8v with Coral vs using Ipcam Custom models with non coral system. Getting ip-cam custom models support with Coral would probably fix this.

I have very good accuracy with the standard models under Yolo 5, so the issue appears to me to be possibly with the Yolo 8 engine. I've only had 1 false positive in 18 months with Yolo 5 caused by a sun shadow, and most of my human triggers are from the waist up only as there's a wall separating them from the camera. Yolo 8 should in theory give much better accuracy as usually each Yolo model claims improvements. However, as it's new it maybe it just needs some more tweeking by the Yolo developers.

 

[tigerwillow1]

Question for those of you who have set your blue iris service to delayed start. Mine takes a bit over 2 minutes to start after the machine boots. Is that typical and/or normal?

 

[Mad_Marik]

Do we have any idea what a camera count would be (generally) for a AI server? Meaning, (40) camera per server, etc.

Using a mid to high box, say r7 with 32gb and a SSD. I mean, I have one running (15) cams currently, but I am about to install a 100ish cam system and was curious if anyone else had done a larger install with a CPAI standalone server.

Thanks a ton.

 

[actran]

@mad-maks Use CP.AI Explorer (Benchmark tab) to get an idea of how many requests/sec it can handle. See screenshot below.

Think about # of requests (or operations) instead of # of cameras.
If you have 100 cameras but few events, or 50 cameras but twice the # events, the load you put on CP.Ai is the same, right?

 

[Mad_Marik]

@mad-maks Use CP.AI Explorer (Benchmark tab) to get an idea of how many requests/sec it can handle. See screenshot below.

Think about # of requests (or operations) instead of # of cameras.
If you have 100 cameras but few events, or 50 cameras but twice the # events, the load you put on CP.Ai is the same, right?

View attachment 189017

Absolutely, how active the cams are would certainly affect things, I was just curious if anyone was running a larger system and what experience they are having.

I will definitely use the calculator tho, I appreciate the response. Did not realize that was there.

 

[xChristopher]

OS: Ubuntu, CPAI v2.5.6 (codeproject/ai-server:cuda12_2), docker

I tried to get Yolov5.NET v1.9.3 to work using GPU and cannot.
I have CPAI setup in a docker container in Ubuntu, I have the nvidia drivers installed, nvidia-runtime, etc.
I enabled Yolov5.NET but it says CPU (even after the first detection) and the inference time is what I would expect from CPU around 200ms.

I do have Yolov5.6.2 working correctly with GPU enabled (inference of about 15ms).
Although, none of Mike's custom models work well for me.
The objects being detected are mostly incorrect (dog as cat, deck as bench).
I have to use Default Object Detection, at which point, it works extremely well for the most part.

I also noticed some of the Environment variables (at least when used with docker compose) explained in their documents just don't work.
For example, this works:

environment:
      - Modules:FaceProcessing:LaunchSettings:AutoStart=false

This does not work:

environment:
      - Modules:ObjectDetectionNet:LaunchSettings:AutoStart=false

Just my insights, not sure there is anything here that can be fixed.

 

[Steve_Texas]

(Hopefully) quick question about configuration for BI and CPAI for License Plate Reading:

How do I keep CPAI from trying to run plates for every camera?

Some context and setup:

For my setup, I am running BI with 9 cameras, one of which is a dedicated LPR cam, on an older Dell server (R710). My CPAI machine is a Windows 10 machine with an i7 and a 1060 Nvidia GPU. I am running the latest version of each (5.8.8.8 x64 and 2.5.6).

For the sake of cleanliness, I fully uninstalled an reinstalled CPAI on the CPAI machine and installed the Object Detection (Yolov5 6.2) and License Plate Reader modules.

For the AI configuration in the "General" settings for BI, I have the following:

My LPR camera is setup this way for AI:


I have noticed that if I check "ALPR for Plates", every camera checks for plates even if their AI configuration is like this:



Is it possible to make only a single camera call CPAI for license plates?

 

[wittaj]

You need to add:

alpr:0

To the custom models box to turn off alpr.


But keep in mind that a recent change to BI now has it run thru all models.

From this post in the BI update thread:

So I was using the following settings for my LPR camera:

I sent this to Ken and this was his response:

Yes with these settings, all custom models will be run. If you want only specific models, you need to specify the model name. If you use a non existent mode name, none of them will be executed.

There may have been a fix made to this code to cause this "new behavior" but it's actually now correct.

Thanks

Ken

So from version 5.8.8.3 on you have to enter a non existent model name in the custom model field if you only want to use the ALPR model. I changed mine to "nothing,object:0" and just updated to 5.8.8.8 and everything seems to be working as expected.

 

[Steve_Texas]

You need to add:

alpr:0

To the custom models box to turn off alpr.


But keep in mind that a recent change to BI now has it run thru all models.

From this post in the BI update thread:

So I was using the following settings for my LPR camera:

I sent this to Ken and this was his response:

Yes with these settings, all custom models will be run. If you want only specific models, you need to specify the model name. If you use a non existent mode name, none of them will be executed.

There may have been a fix made to this code to cause this "new behavior" but it's actually now correct.

Thanks

Ken

So from version 5.8.8.3 on you have to enter a non existent model name in the custom model field if you only want to use the ALPR model. I changed mine to "nothing,object:0" and just updated to 5.8.8.8 and everything seems to be working as expected.

Thanks @wittaj !

 

[sewington]

OS: Ubuntu, CPAI v2.5.6 (codeproject/ai-server:cuda12_2), docker

I tried to get Yolov5.NET v1.9.3 to work using GPU and cannot.
I have CPAI setup in a docker container in Ubuntu, I have the nvidia drivers installed, nvidia-runtime, etc.
I enabled Yolov5.NET but it says CPU (even after the first detection) and the inference time is what I would expect from CPU around 200ms.

Could you share your System Info tab from the CodeProject.AI Server dashboard, and any error logs you see (if any)? Also curious to see the results if you run the nvidia-smi command and then the nvcc --version command.

 

[Pentagano]

Has anyone got the nesh option working properly?

I have CPAI running on my unraid server as the docker app. (UDP and TCP enabled)

2nd CPAI on another pc (proxmox windows VM) (only tcp as I do not know how to add udp on the windows installation). tried to alter the json but it broke).
3rd CPAI on a proxmox lxc container (UDP and TCP enabled)

Only the 3rd option CPAI manages to ping and pick up the 2nd server.
All the others are unresponsive.

My main CPAI is on the unraid server which shows the others in mesh but fails to ping any. Must be a setting on the unraid.

 

[Pentagano]

Given up on the mesh. My CPAI container on unraid will not communicate with the containers on proxmox.
The only visible instance of CPAI in the mesh is the windows 10 installation.
Tried creating new macvlan networks on the proxmox containers but half worked and half broke.

Also I noticed if you use the AITool and mesh is turned on the AITool connection is refused from CPAI.

 

[Vettester]

Given up on the mesh.

It works well with two Windows 10 installations.

 

[Pentagano]

It works well with two Windows 10 installations.

View attachment 190269

Yes I had no issues with windows.
It's the proxmox that is problematic unfortunately.
Only use windows for my work and personal laptops.
All my servers either have unraid or proxmox. Might try a few windows vms bare minimum with CPAI.

Well for sure you cannot use the mesh and the aitool with multiple severs together.
They refuse the connection.

 

[swilliams76360]

Question on Background Remover module. Does it proactively remove the background for all snapshots sent to CPAI from Bi or do I manually have to remove the background from a snapshot and save that as a different file?

 

[VideoDad]

Question on Background Remover module. Does it proactively remove the background for all snapshots sent to CPAI from Bi or do I manually have to remove the background from a snapshot and save that as a different file?

I don't believe that BI has any reason to make calls to the Background Remover module.

 

[digger11]

Hey @MikeLud1 Any idea when a YOLOv8 ipcam-animal custom model might be available?
I've been wanting to give Object Detection (YOLOv8) a try, but am currently using the ipcam-animal custom model on a few of my cameras.

 

[VideoDad]

Hey @MikeLud1 Any idea when a YOLOv8 ipcam-animal custom model might be available?
I've been wanting to give Object Detection (YOLOv8) a try, but am currently using the ipcam-animal custom model on a few of my cameras.

I have the same question for ipcam-combined.

 

[MikeLud1]

@digger11 & @VideoDad I have been extremely busy with my day job so I have not had much time to work on update my models and adding YOLOv8 models. My day job is starting to slowdown so in a month or so I will have more time to work on the models.