5.4.9 - June 25, 2021

Did some reconfiguration last night after installing a Quadro P400 in my BI PC. I ran into some weird issues and had to make some interesting changes.

I disabled DS, installed the P400 and requisite Nvidia drivers. After this, CPU kept running well over 90% without making any other changes. I eventually found that I had to change all cameras that used H265 to H264 and that brought utilization back down to previous levels, around 40% (using Intel QSV HA). I tried using nVidia for HA, but it would only enable HA for half the cameras, so I left it on Intel.

I then installed DS for GPU and re-enabled DS with custom (dark) model.

Before I had the GPU installed, BI CPU would regularly hit 99% CPU during DS runs and return DS errors. With the GPU installed, it hits 80% during DS analysis (with dark model enabled) and no DS errors.

This is running the latest version, 5.4.9.11
 
No change in that regard here.
And you're using Continuous+Triggered in the Record tab with a pretrigger buffer of ?
And is the pre-trigger long enough to get to the next i-frame?

Thank you all for the help! I've made the error of using continuous+alerts vs continuous+triggered. Switching those affected cams to +triggered has brought back pre-trigger main stream recording. I was fairly sure main stream pre-trigger buffer recording also worked with +alerts but I appear to be wrong. Would be a great addition however, if possible.

Using a five second buffer with 15:15
 
Update 5.4.9.11 seems to be screwing with my leading image in the DS analysis. With motion-leading image turned off, I have no t-0, and instead, I'm getting an earlier leading image presumably at the keyframe. Even weirder, with motion-leading images turned on, I'm still getting no t-0, but TWO that are earlier than that, and in reverse chronological order.
I noticed that with the 5.4.9.11 update my DS timing is one analysis every 750ms not every 1 sec as I had it, I think the update changed that. Maybe check that
 
  • Like
Reactions: sebastiantombs
@looney2ns If you want to try one of the more recent updates with that memory problem you're having use 5.4.9.10. .11 and .12 are not handling memory as well as .10 does.
 
I noticed that with the 5.4.9.11 update my DS timing is one analysis every 750ms not every 1 sec as I had it, I think the update changed that. Maybe check that
Thanks but my T+ snapshots are all right around 500ms intervals where I set them. It's just the T-0 and earlier that are suddenly screwy, and still so with 5.4.9.12. I'm not going to scrutinize it too much until I have my upgrade camera tomorrow whose keyframes are greater than 0.25.
 
  • Like
Reactions: sebastiantombs
.12 seemed pretty good as well. I just update to .13 and watched memory wander up and down a little but it'll take some time until the verdict comes in.
 
  • Like
Reactions: looney2ns
It brings up the frame that DS used for successful detection rather than play the entire clip.
 
Start from the top down in the help file. After the boiler plate the recent features are described.
 
Clip playback, especially multiple back-to-back clips has improved.
 
I assume, and we know what that does, that sets the maximum time that DS will search in models attempting to make a detection.
 
  • Like
Reactions: 105437
Let us know what you find out. I have a thing about Zoom and refuse to use it.
 
Man I was hoping upgrading cameras would fix this, but I'm still having an issue where my leading image of every DS analysis is at T-4500ms or so. And then there is no T-0, so the second image is my T+500ms. Could a setting cause this, or did is this a bug? Nobody else is seeing this? I even tried reinstalling DS, but I fear it's a bug where maybe I would have to reinstall BI.
 
re you running the CPU version and if so what processor and how much memory?