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/

Excited to move to SenseAI and I am technical. But really want to make the move once they are out of their beta phase. For me beta is when I have to manipulate code and or config files in order to change behavior. Not a big deal (at least for me) but it is a sign that they are still on their roadmap for a 1.X release. But I am excited for this as DS has shown memory issues with my CPU setup. Following this thread!
 
  • Like
Reactions: sebastiantombs
Excited to move to SenseAI and I am technical. But really want to make the move once they are out of their beta phase. For me beta is when I have to manipulate code and or config files in order to change behavior. Not a big deal (at least for me) but it is a sign that they are still on their roadmap for a 1.X release. But I am excited for this as DS has shown memory issues with my CPU setup. Following this thread!
Hi to clarify there is no need to modify any config file, people chose to because they prefer that approach for Docker. I however don't use that approach, doesn't mean I am wrong\correct or better just different.
Deepstack is very much beta, so that's not an argument.

If it helps, this is how I run without any config file modification at all to use custom models (custom models are OPTIONAL)

a) Create a folder called: "C:\ProgramData\CodeProject\SenseAI\models"
d) Download your custom models (eg: IPcam-general.pt) and place in this folder.

To create the docker container with the options from powershell I run;
docker run -d -p 5000:5000 --name 'SenseAI-Server' --restart always -e VISION-SCENE=False -e VISION-FACE=False -e VISION-GENERAL=True -e MODELSTORE-DETECTION=True -e Mode=Medium -e Profile=desktop_cpu -e MODELS_DIR=/usr/share/CodeProject/SenseAI/models -v C:\ProgramData\CodeProject\SenseAI:/usr/share/CodeProject/SenseAI codeproject/ai-server
(You can read the docker manual, but -e passes through an environment variable, the -v is where I mount the local volume for use by the container. The environment variables are not well documented, but adjust to your needs.

d) Configure BlueIris as you would to point to the same custom model folder, and reference "ipcam-general" in the AI triggers as shown below;

PS: I do realise that my confirm labels may not all exist in the general file. Doc said general also includes dark models. I will also test the combined model and wee where I land.
1659055277324.png
1659053449438.png
 
Last edited:
Thanks for the info. I use Docker/Rancher at work. I deploy Python and Java on a regular basis to Docker. Python was tricky to say the least. Love it!
I liked Deepstack as I did not have to spin up a container. I just pulled down some files, pointed BI to the exe and off it went. I am pretty sure that the steps you lined out above will be scripted at some point.
So for sure moving to SenseAI and it is a given. It seems to have improvements in performance, accuracy, cpu and memory usage which is good.
Having fun watching it move into the main stream by innovators such as yourself!
 
a) Create a folder called: "C:\ProgramData\CodeProject\SenseAI\models"
d) Download the "custom" model general.pt and re-name it ipcam-general.pt (no idea why, but SenseAI was specifically looking for this file)

As previously mentioned the custom models are already included in the install and are located in the "C:\Program Files\CodeProject\AI\AnalysisLayer\CustomDetection\assets" directory so there is no need to download them separately.
 
  • Like
Reactions: DanDenver
As previously mentioned the custom models are already included in the install and are located in the "C:\Program Files\CodeProject\AI\AnalysisLayer\CustomDetection\assets" directory so there is no need to download them separately.
I was clearly referring to a docker install so that path does not exist. Your post made me dive deeper, and I was able to locate the assets in the docker image. So yes no need to download these unless BI needs these. Earlier poster said BI needs these.
/app/AnalysisLayer/CustomDetection/assets/IPcam-combined.pt
/app/AnalysisLayer/CustomDetection/assets/IPcam-dark.pt
/app/AnalysisLayer/CustomDetection/assets/ipcam-general.pt
/app/AnalysisLayer/CustomDetection/assets/license-plate.pt
/app/AnalysisLayer/CustomDetection/assets/IPcam-animal.pt

So the revised command for our docker friends who do not wish to use a config file would be;
docker run -d -p 5000:5000 --name 'SenseAI-Server' --restart always -e VISION-SCENE=False -e VISION-FACE=False -e VISION-GENERAL=True -e MODELSTORE-DETECTION=True -e Mode=Medium -e Profile=desktop_cpu -e MODELS_DIR=/app/AnalysisLayer/CustomDetection/assets/ -v C:\ProgramData\CodeProject\SenseAI:/usr/share/CodeProject/SenseAI codeproject/ai-server

Update: Re-tested, for BlueIris to realise the models exist, I still needed to download the custom models and point BlueIris to them.
 
Last edited:
Thanks for the info. I use Docker/Rancher at work. I deploy Python and Java on a regular basis to Docker. Python was tricky to say the least. Love it!
I liked Deepstack as I did not have to spin up a container. I just pulled down some files, pointed BI to the exe and off it went. I am pretty sure that the steps you lined out above will be scripted at some point.
So for sure moving to SenseAI and it is a given. It seems to have improvements in performance, accuracy, cpu and memory usage which is good.
Having fun watching it move into the main stream by innovators such as yourself!
Sorry but I don't think you get how Docker works. There is nothing to script, as Vettester pointed out the first two steps were not required, so it's literally a single command line, stating the options you require. That single command downloads and installs and runs the container. Can you really get any simpler?
No additional BI config than you needed for Deepstack.
 
  • Like
Reactions: DanDenver
I stumbled across this today and it may explain why tables and trees can be IDed as "people" and so on. I have zero idea what models have been used for initial training for SenseAI or DeepStack, but it's not much of a stretch to think these models are involved one way or the other.

 
I stumbled across this today and it may explain why tables and trees can be IDed as "people" and so on. I have zero idea what models have been used for initial training for SenseAI or DeepStack, but it's not much of a stretch to think these models are involved one way or the other.

I just read the article and they did not look at the COCO dataset for errors.

The dataset (images) used for training DeepStack & CodeProject.AI is the COCO Dataset.

DeepStack uses a 2 year old version of yolov5 V3.0 for its models

The current version of CodeProject.AI is using DeepStack. The next version of CodeProject.AI will be using the code from the current release of yolov5 v6.1

Here is a research paper on the COCO Dataset
 
Last edited:
CodeProject.AI update.
I have been working with CodeProject.AI team to have my custom models included with the install. The next release should have my models included. Also they added the ability to benchmark the models and more logging details. They still are working on GPU support, they had some Windows install issues that they needed to resolve first. The version I tested this morning looks to resolve those issues.


View attachment 133480
How do I get to a page like the one in the attachment? Thanks.
 
I have 3 different servers. One of them worked, the other two did not. What do I need to do? I uninstalled DeepStack. Do I need to uninstall the DeepStack folder in C?
 
In my auto start/stop with blue iris, it still lists DeepStack. How do I get this changed like I see others have? Code ProjectAI Service
I had to modify the windows registry to get mine to work. You may want to look at this post:
 
I had to update my BI to higher than the known stable one. I'm sure it is stable, but it wouldn't work with the 5.5.8.2 version. So I went with the 9.3 version. It worked, so I guess I didn't read well enough or something to have missed that it needed something higher than 8.2.
 
I had to modify the windows registry to get mine to work. You may want to look at this post:

I changed by RegEdit (as admin) but it reverts back after a reboot; 'Use custom model' still greyed out.

Wonder what's going on?

Running BI v5.5.9.5 and v1.5.5.1 of SenseAI
 

Attachments

  • Capture.PNG
    Capture.PNG
    13.5 KB · Views: 38