Deepstack declaring wrong objects ?

[euquiq]

Hello!

I am searching for "person" objects on my security camera. Deepstack just returned the following two "person" objects:

Is there any further information you would need for investigating this ? I.E.: I can try to find the original snapshot from my video storage and send it to you so you can further analyze locally how did DeepStack ended up labelling this as two "person" objects ?

[AbimbolaOO]

@euquiq Deepstack actually did not detect two people but rather it detected a "human being" and a "bicycle". It means the bounding boxes are for the human and bicycle respectively. You might need to explicitly inform deepstack to detect just humans.

[euquiq]

@AbimbolaOO As I stated, Deepstack declares that both objects are "human beings"

[johnolafenwa]

Hello @euquiq Can you share the full json response from DeepStack including the predicted labels?

[euquiq]

Sorry, no, I did erase it. The software we made parsed that json, and boxing in green al objects of type "person" (it's a condition for the boxing to occur, that Deepstack labels the object as "person").

I think I stored the still images that caused this: The last image is the one that triggered the "person person" response from Deepstack. in case you want to input these into Deepstack (it was configured in high).

Also when calling deepstack, if I correctly recall, the parameter I can pass thru is "minimum confidence". Is there a way to also limit Deepstack to "person" objects (and would this accelerate deepstack process ?)

Thanks!