DeepStack not analyzing after a trigger....why?

[Sybertiger]

I must have a mental block or something. When a cam triggers due to the BI5 motion sensor trigger, DS is not doing the analysis for one of my cams. What am I overlooking that I haven't overlooked on the other 10 cams that DS is doing the analysis after a trigger?

<<< SEE, there is no DS analysis happening after this trigger

 

[Sybertiger]

See...two cams captured the same motion using BI motion alerts. One cam the DS analysis is done but the problem cam no DS analysis.

 

[wittaj]

Has this always happened or did it just recently stop working?

What version BI are you on?

 

[Sybertiger]

Which is the very last release before SenseAI was brought onboard.

It has always happened. This is a clone of a PTZ cam but done exactly in the same way another cam (non-PTZ) was cloned. Motion triggers work exactly the same way but in this case DS is not doing the analysis. Must be something stupid I've overlooked.

 

[Sybertiger]

BTW, the PTZ is stationary at preset #1 so it's not moving around at all.

 

[wittaj]

I have noticed some wonky things sometimes with clones that don't exactly work as planned.

For kicks, put DS on the Master PTZ cam instead of the clone and see if it does it.

 

[sebastiantombs]

How about click a check mark on "motion triggers only" in the AI tab?

I use AI on clones almost exclusively and have no problems at all. Kind of strange to me.

 

[Sybertiger]

How about click a check mark on "motion triggers only" in the AI tab?

I use AI on clones almost exclusively and have no problems at all. Kind of strange to me.

I've tried that with no success along with a million other combination of things. The only thing I came up with since is probably to delete the clone and start over....LOL.

Gonna add the motion detect back to the PTZ and if it works clone it from there.

 

[Sybertiger]

Maybe there is something willy-nilly about the DS integration and perhaps the reason for going to SenseAI. For the cam I mentioned that picks up the motion then sends it to DS for analysis, DS still is having issues with black colored vehicles. What's odd is when you use Testing & Tuning > Analyze with DeepStack the black vehicles easily are detected by DS all day long.

 

[wittaj]

Nope, not the reason.

Keep in mind that the "Analyze with Deepstack" under "Testing & Tuning" will ALWAYS perform better than live as it is after the fact and should not be used as an analysis tool to try to figure out why it didn't see and trigger for a car or person. It should only be used to see what DeepStack can find in that clip, like "hmm I wonder if DeepStack can find a toothbrush" and then walk around with a toothbrush and have it identify it. I can run this on a camera not using Deepstack and it will show EVERYTHING that Deepstack has in its objects to find that it sees in the clip. This method will show you EVERY ITEM deepstack searches for.

It would do the same thing with BI motion or SenseAI if you ran it thru Testing and Tuning and yet it wasn't analyzed live.

 

[Sybertiger]

Understood, but not understanding what or what not is going on...LOL.

This baffles me. Nothing was found....or was it?

 

[sebastiantombs]

Keep us posted. This is a real head scratcher. I do wonder if it's a unique problem related to the specific camera and how BI sees it rather than a DS problem.

 

[Sybertiger]

So, what caused DS to run the general model only one time. See below, general is used for the analysis one time then you don't see the general model used in subsequent runs.

 

[sebastiantombs]

To be fair, the capture is pretty well hidden behind that bush. I've cut the pre-trigger down to zero on a few cameras just to avoid problems like that.

 

[Sybertiger]

Let me rephrase....I wasn't concerned that DS could not find a vehicle behind a bush. I'm concerned that DS isn't running the GENERAL model but one time out of SIX tries. The GENERAL model is the only model that can find a car or truck.

 

[MikeLud1]

Let me rephrase....I wasn't concerned that DS could not find a vehicle behind a bush. I'm concerned that DS isn't running the GENERAL model but one time out of SIX tries. The GENERAL model is the only model that can find a car or truck.

It might be that you are using the CPU version of DeepStack and running 3 models is to much work for the CPU, try just running the general model by its self to see if it makes a difference.

 

[integrity]

How many instances of DeepStack are you running? You're queueing up ~22 seconds of CPU time in 1,2 seconds so it might be some kind of buffering issue. You have 8 threads so I'd be running 6 instances - one per analysis leaving two threads free.