Just one more before I shut up for a while. It may be that 5.4.9.1 isn't good for me or else all my other tweaks are now the problem. To confirm how the system works, does BI decide to compile an image to send to DS and when the pre-trigger starts, ending up being1.4 seconds or probably more too early this time? (pre-trigger buffer 4 seconds)
I've tweaked, tinkered, adjusted and crossed my fingers regularly. If only DS worked real time like it does in "tuning" it would be an awesome addition. It does eliminate a lot for false triggers for me, but still isn't consistent enough to be relied on. I've finally gotten the dark model working and it is detecting things better at night, but still not totally reliable. I think we have to be patient and give Ken/BI a little time to fine tune things, either that or the only other thing I can think of is add DS in a Linux container or box and run it there. I'm not a Linux fan, but that may change with Win11 coming out and support for Win10 ending. Even that isn't set in stone because if Win10 works for BI who needs the crappy updates MS puts out. None of my machines update unless I let them. Maybe it is time for me to look, seriously at a Linux box for DS.
I wouldn't hold your breath for end of support. Windows 7 only reached end of support January last year, five years after the release of WIndows 10, which wasn't even the first OS to be released after it. Windows 10 should remain in support at least until the middle of the decade.
I wouldn't hold your breath for end of support. Windows 7 only reached end of support January last year, five years after the release of WIndows 10, which wasn't even the first OS to be released after it. Windows 10 should remain in support at least until the middle of the decade.
I've tweaked, tinkered, adjusted and crossed my fingers regularly. If only DS worked real time like it does in "tuning" it would be an awesome addition. It does eliminate a lot for false triggers for me, but still isn't consistent enough to be relied on. I've finally gotten the dark model working and it is detecting things better at night, but still not totally reliable. I think we have to be patient and give Ken/BI a little time to fine tune things, either that or the only other thing I can think of is add DS in a Linux container or box and run it there. I'm not a Linux fan, but that may change with Win11 coming out and support for Win10 ending. Even that isn't set in stone because if Win10 works for BI who needs the crappy updates MS puts out. None of my machines update unless I let them. Maybe it is time for me to look, seriously at a Linux box for DS.
If you PC hardware supports it you could try installing WSL2, Windows Subsystem for Linux which would allow you to run Docker, then use Docker to run a Linux instance of Deepstack.
This is how I ran Deepstack when I was using AITools which had much performance than running the Windows version of Deepstack.
This article sort of explains how to get Docker going, you just then need to pull down DS and run it within Docker.
@IAmATeaf As soon as I saw "Install Windows Insider Preview build 18975 or later" that idea is kind of nixxed. I've got too much time and effort on the install of Pro that runs on my BI system, not to mention the configuration and tuning of BI itself, to start all over again. Good suggestion though. I jumped through hoops to get the GPU version running and had all kinds of problems with "NVIDIA container" running multiple instances and taking the CPU to its knees in the process. That turned into a headache every time I rebooted the machine for any reason at all so I'm running the CPU version. Maybe that's the problem. Beats me. As long as 5.4.8.2 is running fine I'm happy enough.
I just looked at the version on my laptop and it's 18362. I've had updates shut off that long on this one and the BI machine has updates shut off only a day or to later than this one. Maybe, maybe, if all else fails I'll give it a stab.
On question though, how has the 5.4.9.x series worked for you?
And I'm guessing you're using a docker container for DS. I just gave 5.4.9.4 a quick try. It worked fine for one detection, then started "nothing found" and "deepstack: 100" again. Back to 5.4.8.2 again.
And I'm guessing you're using a docker container for DS. I just gave 5.4.9.4 a quick try. It worked fine for one detection, then started "nothing found" and "deepstack: 100" again. Back to 5.4.8.2 again.
As I said before when I used to use DS with AITOOLS I used to run DS within docker using WSL2.
For me DS was fine during the day but very poor at night so gave up on it a few weeks later and moved over to using IVS in the cams themselves which I found to be way more reliable.
Any ideas why this showed as cancelled when clearly the .dat file shows that it recognized a person? My confidence is set to 30%. Does it cancel because it sees other items in the same frame?
Any ideas why this showed as cancelled when clearly the .dat file shows that it recognized a person? My confidence is set to 30%. Does it cancel because it sees other items in the same frame?
Is the analysis you show here a product of you streaming video with 'Analyze Using DS' enabled (from the right click menu of a streaming video) and watching the results as they occur (not optimal)? Or did you drag and drop the applicable .dat file from the Alerts folder into the DS tab of the status window (best option)?
If there was any type of error in the frame(s) before your 'person 66%' to that then confirmation won't happen and the alert would be cancelled--that could be one explanation. Did you see an error 100 below the person 66% you have highlighted?
If for some reason the alert was cancelled prior to the highlight, then subsequent frames won't reverse the cancellation. Was there an asterisk after any of the analyzed frames prior to the highlight? Did you get an asterisk at all?
Is the analysis you show here a product of you streaming video with 'Analyze Using DS' enabled (from the right click menu of a streaming video) and watching the results as they occur (not optimal)? Or did you drag and drop the applicable .dat file from the Alerts folder into the DS tab of the status window (best option)?
If there was any type of error in the frame(s) before your 'person 66%' to that then confirmation won't happen and the alert would be cancelled--that could be one explanation. Did you see an error 100 below the person 66% you have highlighted?
If for some reason the alert was cancelled prior to the highlight, then subsequent frames won't reverse the cancellation. Was there an asterisk after any of the analyzed frames prior to the highlight? Did you get an asterisk at all?
The asterisk is a confirmation that the alert is either confirmed or cancelled--in this case cancelled.
In your example, when you scroll down through the analysis you'll likely not find 'person 66%' or any other object you've designated in the To Confirm box of the camera's AI settings. I'm guessing you've also designated a number of additional images to analyze at 750 msec? I can't explain how the person was missed with this setup, but ensuring your AB motion zone rules are adequate, reducing CPU usage while analysis is going on (select Low or Medium in the AI tab of BI options), and consider not using facial recognition, which as of now seems to cause more problems than it's worth.
/cdn.vox-cdn.com/uploads/chorus_asset/file/16292558/windowslogo_1.jpg)
