IP Cam Talk Custom Community DeepStack Model

[MikeLud1]

I'd say "car" and "Car" or even "car" and "Vehicle". That way you could add trucks as well and just use a single label.

In the dark model there are no trucks.

Dark	Total Images	Train Images	Val Images	Image Total
Bicycle	652​	624​	28​	652​
Bus	526​	504​	22​	526​
Car	637​	610​	27​	637​
Cat	727​	696​	31​	727​
Dog	796​	762​	34​	796​
Motorbike	502​	481​	21​	502​
People	608​	582​	26​	608​

 

[105437]

I vote for uppercase for the objects for the dark model.

 

[MikeLud1]

I could also do (CarDark, CatDark, & ....)

 

[sebastiantombs]

That would work, but would mean a pretty small test display in alerts. Upper/lower case is probably better in that regard.

 

[MikeLud1]

That would work, but would mean a pretty small test display in alerts. Upper/lower case is probably better in that regard.

I am going with Upper/lower case

 

[Jason R]

Since installing these custom models, DS is only detecting people. The original model was detecting cars and dogs, but that's not being identified anymore (even with a reduced certainty).

 

[MikeLud1]

Since installing these custom models, DS is only detecting people. The original model was detecting cars and dogs, but that's not being identified anymore (even with a reduced certainty).

Which model are you using?

 

[Jason R]

Which model are you using?

Pulled it from Github (GitHub - MikeLud/DeepStack-Security-Camera-Models) yesterday

 

[MikeLud1]

Pulled it from Github (GitHub - MikeLud/DeepStack-Security-Camera-Models) yesterday

For (person, bicycle, car, motorcycle, bus, truck, bird, cat, dog, horse, sheep, cow, bear, deer, rabbit, raccoon, fox, skunk, squirrel, pig) detection you should only use combined.pt

 

[MikeLud1]

Pulled it from Github (GitHub - MikeLud/DeepStack-Security-Camera-Models) yesterday

Your camera AI setting should look like this

 

[Jason R]

Your camera AI setting should look like this
View attachment 113257

I'm running DS on a docker on a different computer, so I assume my "custom models" field will be blank on BI?

 

[MikeLud1]

I'm running DS on a docker on a different computer, so I assume my "custom models" field will be blank on BI?

You still need

 

[Jason R]

You still need
View attachment 113264

Ok. For my own education, what does this do?

In the "to confirm" field, I assume I can also have all the other labels in your model, like pig, deer, etc..

 

[wittaj]

It won't run a custom model if you don't tell it to.

The objects:0 tells it not to run the default object detection.

 

[Jason R]

It won't run a custom model if you don't tell it to.

The objects:0 tells it not to run the default object detection.

Interesting. I would then think with this field blank I'm still running the default DS model then - but since I installed these models it's no longer detecting dogs and cars, so it seems like the original model performance is worse now...

 

[wittaj]

Did you place the custom models in a different directory? If you stuck them in the main DS folder and not a subfolder then it won't work.

 

[wittaj]

What are your times for nothing found and any .DAT file analysis to help troubleshoot?

 

[Jason R]

What are your times for nothing found and any .DAT file analysis to help troubleshoot?

The custom models are in a subdirectory.
The average time is ~ 400ms regardless if anything found or not.
Where would I expect to find .DAT files?

 

[wittaj]

Do you have checked the box on the screenshot above "Save Deepstack Analysis Details"

If not, check it and then wait for a trigger.

 

[Jason R]

Do you have checked the box on the screenshot above "Save Deepstack Analysis Details"

If not, check it and then wait for a trigger.

Will it save it on the docker side or the blue iris side?