[tool] [tutorial] Free AI Person Detection for Blue Iris

I couldn't get deepstack to run on my windows server 2016 BI machine without python crashing so I set it up in a hyper-V ubuntu docker
I think you may need to increase your BI pre trigger video buffer higher so when it finally does trigger, you can catch it. Or else figure out how to get it to process faster (add more cores or ram to hyper-v maybe?). My average is less than a second on threadripper 16 core machine.
 
  • Like
Reactions: meissen
I was able to cut the time in half to 12 seconds just by throwing more memory at the hyper-V machine; now that it's the weekend I'm going to try and get DeepStack running natively right on Windows instead of having to go through the Hyper-V and docker.
 
Is it possible to use GPU for Deepstack within Docker?

And what is the maximum resolution for fast AI parsing? I guess 4K is a bit too much? Are you all using the substream 640x480 stream for Deepstack AI?
 
Has anyone actually figured out what the fix is for Windows Deep Stack instances when python crashes immediately upon start-up? Majority of the posts I can find about it people just switch to Docker and call it fixed -- someone out there has to have been able to fix the Windows python crash issue?
 
Has anyone actually figured out what the fix is for Windows Deep Stack instances when python crashes immediately upon start-up? Majority of the posts I can find about it people just switch to Docker and call it fixed -- someone out there has to have been able to fix the Windows python crash issue?

DeepQuest for Windows will consume a lot of CPU which is the reason for the suggestion to use docker desktop for windows with WSL2 and the docker version of DeepQuest.
 
  • Like
Reactions: Cameraguy
Has anyone actually figured out what the fix is for Windows Deep Stack instances when python crashes immediately upon start-up? Majority of the posts I can find about it people just switch to Docker and call it fixed -- someone out there has to have been able to fix the Windows python crash issue?
Yes, it was happening to me. Believe I had to clean out C:\Users\username\.deepstack folder. Uninstalling, deleting C:\Deepstack did nothing for me. Something in that folder was corrupt.
 
  • Like
Reactions: meissen
I guess I have more reading up to do on this Docker Desktop WLS2. I tried to install Docker on my Windows Server 2016 but then when I went to add the DeepStack docker I got an error message about it being for linux only. That's when I then created the Hyper-V ubuntu server virtual machine, installed docker, and then installed the deepstack docker to it. Boosting CPU and memory I was able to bring the DeepStack's processing time down to 3 seconds, I guess that's bearable with the buffer in Blue Iris.

So if I understand the Pre-Trigger Buffer on the Record tab - since my DeepStack is taking ~3 seconds to process and send the trigger command to Blue Iris, should I put the buffer at 4 or so seconds so I always have the alert just before the motion starts?
 
Is it not possible to use Deepstack on Docker in Windows using GPU instead of CPU?

Also, I can't seem to get my BlueIris push notifications working on the BlueIris mobile App. I set up the AI tool to trigger the (HD stream of the) camera in BlueIris. And set up push notifications for IOS.
 
I guess I have more reading up to do on this Docker Desktop WLS2. I tried to install Docker on my Windows Server 2016 but then when I went to add the DeepStack docker I got an error message about it being for linux only. That's when I then created the Hyper-V ubuntu server virtual machine, installed docker, and then installed the deepstack docker to it. Boosting CPU and memory I was able to bring the DeepStack's processing time down to 3 seconds, I guess that's bearable with the buffer in Blue Iris.

So if I understand the Pre-Trigger Buffer on the Record tab - since my DeepStack is taking ~3 seconds to process and send the trigger command to Blue Iris, should I put the buffer at 4 or so seconds so I always have the alert just before the motion starts?

I would add the processing time to what ever your pre-trigger is set to, so for me I had it set to 5, my processing takes around 3.5 so I have it set to I think 7 or 8.
 
  • Like
Reactions: meissen
Thanks! Finally have it really working good for triggering the camera and it capturing the activity on the main stream. Very exciting to finally have true motion alerts instead of the wind blowing in the trees!
 
  • Like
Reactions: Cameraguy
What system do you use for Deepstack?
I have an i5-8500T 2.1 GHz and 8GB of ram where I also run Blue Iris with 4 hikvision cameras.

I send 2048x1536 resolution files to Deepstack which analyse them in around 6-700ms.

So anything over 1 second is very slow to me.
 
I'm running everything on a Dell R510 with dual L5640 2.26ghz and 32GB of ram, the virtual ubuntu machine running docker I put to 10GB for memory and 12 virtual CPUs.
 
Next question - in the instructions it says if you check the checkbox for "Hidden" that it will hide the camera from the UI3 interface, but that's not happening for me? My hidden camera is still showing up in my UI3 interface. Is there an additional flag I have to set somewhere to hide hidden cameras from UI3's interface?
 
On console you have to right click and hide all hidden cameras by unchecking show hidden cameras
 
Last edited:
That only hides it in the actual BI program, but if webcasting is still enabled then it won't hide it in the UI3 web interface.
 
Ok, I have read to page 12 and I am running on little to no sleep so I jumped ahead for a more direct approach for an answer so i apologize if it has been discussed before ;) But I was wondering if the benefits are any better running a windows docker for DeepStack, BI and AITools all on the same system (HP ProDesk 400 G3 SFF, i5-6500, 8GB 2133 DDR4, 3TB 7200 ENT HDD, 120GB SSD) as opposed to running the windows version of DeepStack? I have noticed that there is quite a delay when motion is detected to when the image is processed and posted in AITools. Are there tweaks to make this faster either way? I have only setup 1 camera (AD110 Doorbell) to test so far till I know the ins/outs of how to make this work smoothly.
_TIA for all your help and TY @GentlePumpkin for the great tools and step/step guides!