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/

[gwithers]

I know about the override server option, but how do you define a custom model folder for the second machine?

Sorry, I cut off that in the image but you just specify the custom model you want to apply to that camera in the custom models section. The custom model needs to be on the AI server of course and the default install puts many of the custom models by default. In the example below, the override server at 192.168.4.8 will run its images through the ipcam-combined model.

 

[MikeLud1]

I know about the override server option, but how do you define a custom model folder for the second machine?

One of the integration BI can send a request for the custom models list instead of looking at a folder for the models

 

[Vettester]

One of the integration BI can send a request for the custom models list instead of looking at a folder for the models

Tell me more. How do you configure it to send the request?

 

[gwithers]

Doesn't the AI-server explorer benchmark web page show the installed models in the drop down menu? I apologize if I am misunderstanding the question.

 

[Vettester]

Doesn't the AI-server explorer benchmark web page show the installed models in the drop down menu?

Yes, the installed models are in both drop down menus. However, BI seems to be using only the custom model directory on one machine.



In the screenshot above you can see I have set my custom model folder to use the models on the second instance of CodeProject AI running on 192.168.86.57. Now if I go into the camera's setting and use the override server option to change to 192.168.86.58 BI doesn't find the license plate custom model.

 

[gwithers]

that is Interesting. I have not run into that. In my particular setup, I have the AI models stored locally to BI and they are the same models available in my two VM's running the AI-server docker containers. So the local models folder matches what is on the two remote AI servers. In this scenario, the custom model is used properly on the second AI server as well. Why it is not in you setup is not clear. That is probably going to need some insight from MikeLund1. Would BI perhaps prefer the use of either a fully qualified domain name (FQDN) or an IP address instead of "Home-assistant" netbios/local type name?

 

[Vettester]

that is Interesting. I have not run into that. In my particular setup, I have the AI models stored locally to BI and they are the same models available in my two VM's running the AI-server docker containers. So the local models folder matches what is on the two remote AI servers. In this scenario, the custom model is used properly on the second AI server as well. Why it is not in you setup is not clear. That is probably going to need some insight from MikeLund1. Would BI perhaps prefer the use of either a fully qualified domain name (FQDN) or an IP address instead of "Home-assistant" netbios/local type name?

OK, I'm sorry I even brought this up. I rebooted both my machines and everything seems to be working as expected.

Anyway, thanks for the responses!!

 

[gwithers]

No worries. At least it works!

 

[Tusabrat]

Is there anyone else that is using BI and AI solely for pets, that can share their settings? I am having the most frustrating time. Since upgrading to the latest verion of CPAI (1.6.6-Beta) using CPU, I have lost the triggers for my cats (they were working before, but not reliably. Now there is nothing). The 'person' triggers are working however.

An example: In the lounge, my cat will slowly walk across the room. I have a motion rectangle which follows the cat, so that part is working. However, the trigger is never fired. Yet, when a random pixel flashes on the screen, the trigger is immediately fired. I am so frustrated - why a quick pixel change but not my leisurely strolling cat? (I am using the custom models of ipcam-combined and ipcam-dark. My confirm is cat, dog,Cat,Dog). As I said, while not perfect, it was firing on 1.6.5 but not on 1.6.6. I did roll back, and it got better, but then the lounge camera stop seeing the cats so I upgraded back to 1.6.6 but it didn't change anything).

So if anyone has pet related settings, please share. (I'm not sure if I should be upgrading my cameras (Tapo C110) to try an eliminated the shifting pixels?)

Edit #1 - added screenshots. As you can see, my huge 8kg cat was completely ignored. It didn't even save anything for the Analysis. Here are my settings - I have changed these ad nauseum, including unchecking the Use The Main Stream, and changing the amount of images to process and the time... I have tried using motion with and without zones, with and without the settings such as how many pixels to travel. So I don't understand why I keep getting triggers and alerts on nothing that can be seen, yet my very obvious cat gets totally ignored.

Edit #2 - deselected the custom models and went back to objects - see screenshot of cat labelled as dog. I can live with that, it's better than having nothing. As an aside, why would the object model have things like books and potted plants? It's not like they are useful for triggers and alerts.

 

[Vettester]

I am so frustrated

Can you post a screenshot of the motion sensor configuration from your camera’s trigger page?

 

[MikeLud1]

Is there anyone else that is using BI and AI solely for pets, that can share their settings? I am having the most frustrating time. Since upgrading to the latest verion of CPAI (1.6.6-Beta) using CPU, I have lost the triggers for my cats (they were working before, but not reliably. Now there is nothing). The 'person' triggers are working however.

An example: In the lounge, my cat will slowly walk across the room. I have a motion rectangle which follows the cat, so that part is working. However, the trigger is never fired. Yet, when a random pixel flashes on the screen, the trigger is immediately fired. I am so frustrated - why a quick pixel change but not my leisurely strolling cat? (I am using the custom models of ipcam-combined and ipcam-dark. My confirm is cat, dog,Cat,Dog). As I said, while not perfect, it was firing on 1.6.5 but not on 1.6.6. I did roll back, and it got better, but then the lounge camera stop seeing the cats so I upgraded back to 1.6.6 but it didn't change anything).

So if anyone has pet related settings, please share. (I'm not sure if I should be upgrading my cameras (Tapo C110) to try an eliminated the shifting pixels?)

Edit #1 - added screenshots. As you can see, my huge 8kg cat was completely ignored. It didn't even save anything for the Analysis. Here are my settings - I have changed these ad nauseum, including unchecking the Use The Main Stream, and changing the amount of images to process and the time... I have tried using motion with and without zones, with and without the settings such as how many pixels to travel. So I don't understand why I keep getting triggers and alerts on nothing that can be seen, yet my very obvious cat gets totally ignored.

Edit #2 - deselected the custom models and went back to objects - see screenshot of cat labelled as dog. I can live with that, it's better than having nothing. As an aside, why would the object model have things like books and potted plants? It's not like they are useful for triggers and alerts.

If your only concern is animals (cats & dogs) you should use the ipcam-animal and ipcam-dark models

 

[Tusabrat]

Can you post a screenshot of the motion sensor configuration from your camera’s trigger page?

I'm not sure how helpful it will be as I've tried a matrix of every option, but here is the current setting. As I've said, I've tried with and without zones.

 

[Tusabrat]

If your only concern is animals (cats & dogs) you should use the ipcam-animal and ipcam-dark models

When I tried that, us humans kept causing alerts as we were seen as dogs. Once I added the ignore persons, it got better for a bit. Then it just stopped working altogether.

Here is the latest screenshot using just the objects. My cat, descending from the cat tree has been highlighted as a motion event, but the AI here appears to be too busy labelling static nonsense (which I've said to ignore) to bother with the cat. (He climbed all the way down the cat tree, so a big motion event, but no trigger. Yet random bits of dust and pixels trigger?)

 

[Vettester]

I've tried with and without zones.

Try it again without using zones.

Here are the motion settings I'm using for detecting if our dog is in our living room.





And here are my AI settings.



Note: This is an older camera that doesn't support sub-streams and works better using edge vector algorithm.

 

[Tusabrat]

Try it again without using zones.

Here are the motion settings I'm using for detecting if our dog is in our living room.

View attachment 141835

View attachment 141834

And here are my AI settings.

View attachment 141837

Note: This is an older camera that doesn't support sub-streams and works better using edge vector algorithm.

Ah - your model is yolov5l.... For now, I have uninstalled Codeproject, and at least I am getting plain BI recordings every time my cats are sauntering around. I'll give it a day for my mushroom cloud to subside then try again... thanks for the info.

 

[Corvus85]

Hey guys,

I can't seem to get custom objects working. At the moment, I'm getting around ~750ms for detection, but I'm noticing it's detecting a whole bunch of useless crap, so ideally I want to run the ipcam-combined model. I read that SenseAI already comes with custom models installed, and indeed I can see the 'blank' model files it copied to my C:\DeepStack\MyModels folder, including ipcam-combined.

Here's my Global AI settings:


And here's the settings for the AI in my cameras:


The problem is that even though I've specified the ipcam-combined model in the custom models field, it still looks like it's using the default model.



What am I doing wrong?

 

[jrbeddow]

Hey guys,

I can't seem to get custom objects working. At the moment, I'm getting around ~750ms for detection, but I'm noticing it's detecting a whole bunch of useless crap, so ideally I want to run the ipcam-combined model. I read that SenseAI already comes with custom models installed, and indeed I can see the 'blank' model files it copied to my C:\DeepStack\MyModels folder, including ipcam-combined.

Here's my Global AI settings:
View attachment 141843

And here's the settings for the AI in my cameras:
View attachment 141844

The problem is that even though I've specified the ipcam-combined model in the custom models field, it still looks like it's using the default model.

View attachment 141845

What am I doing wrong?

Uncheck the Global AI checkbox for Default Object Detection.

 

[Corvus85]

Uncheck the Global AI checkbox for Default Object Detection.

Did that. Now it doesn't detect anything at all. The camera triggers, but no alert is generated.

 

[Tusabrat]

Hey guys,

I can't seem to get custom objects working. At the moment, I'm getting around ~750ms for detection, but I'm noticing it's detecting a whole bunch of useless crap, so ideally I want to run the ipcam-combined model. I read that SenseAI already comes with custom models installed, and indeed I can see the 'blank' model files it copied to my C:\DeepStack\MyModels folder, including ipcam-combined.

Here's my Global AI settings:
View attachment 141843

And here's the settings for the AI in my cameras:

The problem is that even though I've specified the ipcam-combined model in the custom models field, it still looks like it's using the default model.


What am I doing wrong?

You need to select the 'use custom model' option and deselect the 'default object detection' option

 

[Corvus85]

You need to select the 'use custom model' option and deselect the 'default object detection' option

Um... I'd like to, but can't you see it's greyed out? I can't select it.