Various arguments in favour of and against running DeepStack on a separate host

Buttan Butt

Buttan Butt

Getting the hang of it
Jun 4, 2017
88
56
Sweden yeah!
Hi there!

I haven't really seen any discussion about the pros and cons of running DeepStack locally (same host as BI runs at) vs running it on a different host in the LAN.

Currently I'm running DeepStack locally but I'm tempted to set it up on a different host as a Docker image. The reason is that I don't want to put the extra load on my Blue Iris server that the DeepStack server would contribute with. However I'm a bit concerned that transferring images for analyzing between two LAN hosts would increase the time from when an image is sent for analysis until the response arrives. I'm not sure how to think here... Any ideas?
 
How long does it take now for Deepstack to analyze an image and get the results out? Transferring an image over the LAN should take something of a few ms or so. I can't image that would add any appreciable time. The main risk (IMO) for not getting an alert is that you have now doubled your chances for a server failure.
 
Last edited:
  • Like
Reactions: Buttan Butt
Many already do that with one of the many AI Tools folks have created using DeepStack as the engine. I think people wanting to run a Docker or another server for AI would be better off using one of these tools than the integrated DeepStack. The DeepStack integrated is to make it simple and easy for the majority of people. If you go to the effort of creating a Docker and what not, you are better off, for now anyway, to use one of the 3rd party apps that has a lot more functionality with it at the moment and was written with Docker in mind rather than the integrated DeepStack version in BI.

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

This program analyzes motion in Blue Iris cameras in real-time using Artificial Intelligence. It notifies you if a human or an object of your choice is detected and can send alert images to Telegram. It is based on DeepQuestAI, a free multipurpose AI that you can install on your own server...
ipcamtalk.com ipcamtalk.com

Yet Another Free Extension for Blue Iris Adding AI Object Dectection/Reduction in False Alarms/Enhanced Notification of Activity/On Guard

I have added a free Windows extension for Blue Iris that uses the DeepStack AI to allow for the identification of objects causing motion alerts. For the lack of anything better I've called it On Guard. This application should dramatically reduce the number of "false alarms". You can detect...
ipcamtalk.com ipcamtalk.com

Last Watch AI - a new platform-independent AI tool with web interface

Introducing Last Watch AI akmolina28/last-watch-ai is yet another open source tool for glueing Blue Iris to Deepstack AI, inspired by gentlepumpkin's AI tools. I created this tool last year to address some of the gaps and frustrations I was experiencing with the other Deepstack tools...
ipcamtalk.com ipcamtalk.com
 
  • Like
Reactions: Buttan Butt
I initially ran Deepstack on the same machine that Blue iris ran (i5-9xxxx series) on. 500-800ms to analyze images. Fantastic! However, CPU hovered in the 60-70% for hours during snowstorms. More cameras relying on Deepstack for AI, the more CPU usage is required. A combo of Deepstack during a storm (snow, rain, or even insects), Blue Iris recording (I record 24/7 for majority of my cameras), and the number of cameras = unhealthy for a single machine. Granted, I had no GPU on my Blue Iris machine to help the CPU (I believe this would of made a difference).
Tried offloading Deepstack onto a Dell NUC using Ubuntu (i7-4770 cpu....NUC CPU's have much less horsepower compared to desktop versions) and the analyzing of images jumped to 5-8 seconds. Not good at all.
This is when I learned to offload AI duties to the Dahua 5442 series cameras I have.

You need to consider the amount of cameras you have AND the horsepower (CPU, RAM, GPU) to analyze Deepstack images.
In my case, I was unhappy with the high CPU usage during storm events with 6 cameras with Deepstack on same Blue Iris machine.
I was unhappy with the 5-8 seconds to analyze images on my NUC during storm events on 6 cameras.

I am very happy with using Dahua AI for the cameras. I bought some additional Dahua 5442's since then. Now, I only have 3 non-AI cameas for outside use that will rely on Deepstack (2x5231's and 1x4231).
 
  • Like
Reactions: Buttan Butt
Holbs said:
I initially ran Deepstack on the same machine that Blue iris ran (i5-9xxxx series) on. 500-800ms to analyze images. Fantastic! However, CPU hovered in the 60-70% for hours during snowstorms. More cameras relying on Deepstack for AI, the more CPU usage is required. A combo of Deepstack during a storm (snow, rain, or even insects), Blue Iris recording (I record 24/7 for majority of my cameras), and the number of cameras = unhealthy for a single machine. Granted, I had no GPU on my Blue Iris machine to help the CPU (I believe this would of made a difference).
Tried offloading Deepstack onto a Dell NUC using Ubuntu (i7-4770 cpu....NUC CPU's have much less horsepower compared to desktop versions) and the analyzing of images jumped to 5-8 seconds. Not good at all.
This is when I learned to offload AI duties to the Dahua 5442 series cameras I have.

You need to consider the amount of cameras you have AND the horsepower (CPU, RAM, GPU) to analyze Deepstack images.
In my case, I was unhappy with the high CPU usage during storm events with 6 cameras with Deepstack on same Blue Iris machine.
I was unhappy with the 5-8 seconds to analyze images on my NUC during storm events on 6 cameras.

I am very happy with using Dahua AI for the cameras. I bought some additional Dahua 5442's since then. Now, I only have 3 non-AI cameas for outside use that will rely on Deepstack (2x5231's and 1x4231).
Click to expand...

I like your strategy @Holbs and even if it will cost me some $ but I'm actually halways through replacing my older cameras with the IPC-T5442TM-AS and IPC-T5442T-ZE models. Thanks for sharing!
 
I recently moved my Blue Iris from a Virtual Machine (ESXI 6.7) to an i5-10400 system that uses about 1/4th the electricity my old XEON server used. I ran Blue Iris in a VM for years and I was a little concerned how it would do on bare metal. Well, it worked so well, I then added my PLEX Stack to the same machine, again to take advantage of the bare metal Intel GPU and again, it worked great. Now that I have Deep Stack running I will see little spikes of CPU activity. At this point, I have cloned my front 4 cameras for photos only, and the CPU activity is more than acceptable to me still. Here is an hour graph from this morning - granted, I do not see a whole lot of activity so your mileage will vary

Tri - 1548.jpg
 
  • Like
Reactions: looney2ns
biggen said:
Hey I thought I told you to use IVS!!! :)
Click to expand...
The whole IPCT came to my rescue! Looney with his night image settings, wittaj with Dahua AI, Biggen to convert me over to IVS from motion, others with home automation help and computer help. Is why I plan on inviting all of IPCT members to my house for a BBQ some day! (of course, have to sign rubber tire waiver)
 
  • Like
  • Haha
Reactions: biggen and sebastiantombs
It really is a great group of folks here helping each other out!

And you have provided great commentary and documentation showing the how and why and the ins and outs of why you need a camera here versus there, and why you need multiple types of cameras and all the stuff people mention here, but seeing it in action with a repeat violator should allow newbies to see that four 2.8mm cams are not going to cut it!

And you have helped plenty with your automation tips and how to stream BI to a TV, etc.
 
  • Like
Reactions: Holbs, biggen and sebastiantombs
You must log in or register to reply here.
Top Bottom