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/

[Dave Lonsdale]

Hmm. I've never had "nothing found" show up in the Confirmed alerts list so I did a quick test just to try to somewhat replicate your findings.
I left "To Confirm:" blank, and added "banana" to "To Cancel:". BI started showing "nothing found" in the Confirmed alerts list when no valid objects were detected. But if you enter only valid label(s) in "To confirm", then "nothing found" will only show up in the Cancelled alerts list, even if you enter non-existing labels in "To cancel:".

Ok cferd5, thanks for doing a test, replicating my findings. I was also able to replicate your settings, with nothing found showing only in the cancelled alerts list when I enter labels in To confirm. For me, though, the disadvantage with populating the To confirm box is that I want to confirm all 21 labels that are available in the ipcam-animal plus ipcam-dark models. Any I miss out end up in the cancelled alerts list.

 

[cferd5]

Ok cferd5, thanks for doing a test, replicating my findings. I was also able to replicate your settings, with nothing found showing only in the cancelled alerts list when I enter labels in To confirm. For me, though, the disadvantage with populating the To confirm box is that I want to confirm all 21 labels that are available in the ipcam-animal plus ipcam-dark models. Any I miss out end up in the cancelled alerts list.

Dave, do you have "Also re-triggers" checked in the Alerts tab? One of my house doors opening into a camera frame would generate a "nothing found" before the person walked out and the alert would never be sent. Since I have "Minimum time between alerts" set to 60 secs., I was usually missing the event entirely. Enabling "Also re-triggers" fixed it for me.

 

[Corvus85]

Does anyone know why this thing won't detect 'cat' at all? I have no less than 5 neighbourhood cats traverse my yard every day/night and none of my cameras have ever picked them up. They don't even detect them as any other object. I thought it was just a quirk of DeepStack, but it's still doing the same with SenseAI.
It detects people and cars just fine.

And before you ask the obvious, a cat will trigger the cams reliably - it's just that 'nothing found' is always the result.

 

[misterfredsr]

Good afternoon everyone.
I just installed my Nvidia 1660 super GPU and my question is do I still need to uninstall everything and install CUDA and cuDNN first or with this new version just install the GPU downloads?
Thanks

 

[JL-F1]

I get cats, but they can't just walk across the frame, they have to stop a sec or two, then start again. If they just walk across the frame I get nothing found, and I have cat:40 in the confirm

 

[jrbeddow]

I do get cats at night (pretty often in fact) with ipcam-dark, but now that you mention it, I don’t see any AI detection during the day with ipcam-combined.

 

[Dave Lonsdale]

Dave, do you have "Also re-triggers" checked in the Alerts tab? One of my house doors opening into a camera frame would generate a "nothing found" before the person walked out and the alert would never be sent. Since I have "Minimum time between alerts" set to 60 secs., I was usually missing the event entirely. Enabling "Also re-triggers" fixed it for me.

Yes, cferd5, I do have Also re-triggers checked. Maybe when I was getting nothing found before the wanted object appeared it was then not checked. Unfortunately I'm old and can't remember.

 

[Dave Lonsdale]

Does anyone know why this thing won't detect 'cat' at all? I have no less than 5 neighbourhood cats traverse my yard every day/night and none of my cameras have ever picked them up. They don't even detect them as any other object. I thought it was just a quirk of DeepStack, but it's still doing the same with SenseAI.
It detects people and cars just fine.

And before you ask the obvious, a cat will trigger the cams reliably - it's just that 'nothing found' is always the result.

Picking up on the remark by jrbeddow, here is an example of the difference between different AI models this afternoon:-



Few things in this world are absolutely perfect.

 

[bjohnrini]

Can someone tell me what I'm doing wrong? I'm getting nothing found in cancelled alerts.
Thanks

Problem corrected by learning to spell.

 

[actran]

@bjohnrini Last screenshot above indicates a typo: "ipcam-cobmined" should be "ipcam-combined".

Also, enable "Save AI analysis details" so you can see what happened for that triggered event.

 

[bjohnrini]

@bjohnrini Last screenshot above indicates a typo "ipcam-cobmined" should be "ipcam-combined". Also, enable "Save AI analysis details" so you can see what happed for that triggered event.

You are correct. I found the mistake and corrected it. That was embarrassing
Thanks!

 

[Corvus85]

Picking up on the remark by jrbeddow, here is an example of the difference between different AI models this afternoon:-

View attachment 142793

Few things in this world are absolutely perfect.

Sure, but I'm not looking for perfect. I'm looking for something that works well enough.
In the example you gave, at least something was detected, albeit incorrectly.

I have never in the past year while running BI had any AI even identify a cat - correctly or incorrectly.
This is regardless of using DeepStack or SenseAI, or regardless of any model I use.

What am I doing wrong?

 

[Cooltiger]

Sure, but I'm not looking for perfect. I'm looking for something that works well enough.
In the example you gave, at least something was detected, albeit incorrectly.

I have never in the past year while running BI had any AI even identify a cat - correctly or incorrectly.
This is regardless of using DeepStack or SenseAI, or regardless of any model I use.

What am I doing wrong?

Curvus
This is on deepstack. Have you set the cameras to high sensitivity?

 

[Corvus85]

This is on deepstack. Have you set the cameras to high sensitivity?

I've set all cameras to the highest sensitivity (even lower object size and lower min contrast than yours), and the AI sensitivity to 'high' months ago, and no luck with either DS or sAI.
I've used the default models, as well as combined and dark too - and nothing.

However, still detects people and cars fine in all combinations.

What's that animal model you're running in your example? Is that included with SenseAI?

 

[MikeLud1]

I've set all cameras to the highest sensitivity (even lower object size and lower min contrast than yours), and the AI sensitivity to 'high' months ago, and no luck with either DS or sAI.
I've used the default models, as well as combined and dark too - and nothing.

However, still detects people and cars fine in all combinations.

What's that animal model you're running in your example? Is that included with SenseAI?

The animal model is included CP.AI

 

[Corvus85]

The animal model is included CP.AI

Ok so I've included ipcam-animal as a model for each camera.
How come now my AI analysis seems to be using all the models?


Here's my AI settings for that camera:


The 'Custom Models' field reads: 'objects:0,ipcam-combined,ipcam-animal'

Why is it using all of the other models?

 

[biggen]

I can't answer those questions but what is "objects:0" for? Is that leftover from DS? Also, what is the cat:35 and dog:35 doing? Are those CP AI specific or DS specific?

In my confirm field, I just have "person,vehicle" and in my custom models box just "ipcam-general".

 

[MikeLud1]

Ok so I've included ipcam-animal as a model for each camera.
How come now my AI analysis seems to be using all the models?

View attachment 142916
Here's my AI settings for that camera:

View attachment 142917
The 'Custom Models' field reads: 'objects:0,ipcam-combined,ipcam-animal'

Why is it using all of the other models?

Where you doing a Testing & Tuning if so this is normal to use all models

 

[Corvus85]

I can't answer those questions but what is "objects:0" for?

Well I was told by another user in another thread that the only way I can get custom models working is to enable 'default object detection' and use 'objects:0' in the custom models field.
It literally wouldn't recognize custom models any other way.

Here's the screenshot of my AI settings:

Also, what is the cat:35 and dog:35 doing?

It overrides the sensitivity for those specific objects.

Where you doing a Testing & Tuning

What do you mean by this?

 

[jrbeddow]

My suggestions: uncheck the Default object detection box (it definitely doesn't need that checked). Also, unless you are also simultaneously experimenting with the facial recognition feature (very much a work in progress), uncheck both of those boxes as well.
The path for custom models that you are using is the "old" standard for DeepStack, and while some workarounds are included when installing CPAI that allow this to work for now, it isn't the real folder where the models are stored. I don't have the exact real path at my fingertips ATM, as I am not in front of my BI machine, but it is under the CodeProject/Ai folder structure if you look for it.

Edit: you can also remove the "objects:0" from each camera's settings page. The way you had it setup originally was telling CPAI (or DeepStack previously) to first USE Default object detection, then overriding that setting by turning it OFF for each camera with the "objects:0" setting.
Also, if you open the CPAI control panel in a browser, check to see which modules are running; once you "lean this down to the essentials" it only needs to show the YOLO Custom models as a running component.