Blue Iris Deepstack vs last year's Deepstack?

Joined
May 1, 2019
Messages
2,215
Reaction score
3,504
Location
Reno, NV
I have a majority of my outdoor cameras being of the Dahua 5442 series doing 100% of the AI duties (I think I helped pay for Andy's Hawaiian vacation). This was not the case around this time last year with some tire slashing events. I was relying on Deepstack AI and a 3rd party program integration for the same 5442's (was unaware of how great Dahua AI was at the time plus I had 5321's too). I would say 87% reliable as compared to Dahua AI of 96% reliable.
All of my current indoor cameras do not have any AI (5231's & 4231's Dahua). Am thinking of going Deepstack once again just for indoor use.
But with all the talk of Blue Iris Deepstack not ready for prime time... why is this? Deepstack was really awesome a year ago for doing what it does. Is the problem Blue Iris <> Deepstack integration?
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,450
Reaction score
47,588
Location
USA
It is really field of view dependent. But it is getting better with every BI update.

The biggest "issue" is Deepstack Integration with BI is intended to be simple - download Deepstack, run the executable, tell BI to use it, and check some boxes. Works great with the right field of view.

The 3rd party platforms require Docker and a bunch of other things that a lot just don't want to mess with. The advantage the 3rd party has over BI/DS integration at the moment is the granular level that you can customize it to, but like I said the BI/DS integration and customization is getting better with every update.

Just think when BI/DS integration first came out, it was simply send a pic to DS for analysis, then came the option to send up to 5 pics at one second intervals, then came a revision to time the timing of the additional pics, then custom models, etc. It is getting there and with tuning can be a great tool.

The DS/BI integration is leaps and bounds better than it was when it first came out.

We tend to see on this forum the issues folks have with it, but with the right field of view, it is spot on. I only bring up the problematic view I am dealing with, but it works without fail on my other camera views.
 
Joined
May 1, 2019
Messages
2,215
Reaction score
3,504
Location
Reno, NV
It is really field of view dependent. But it is getting better with every BI update.

The biggest "issue" is Deepstack Integration with BI is intended to be simple - download Deepstack, run the executable, tell BI to use it, and check some boxes. Works great with the right field of view.

The 3rd party platforms require Docker and a bunch of other things that a lot just don't want to mess with. The advantage the 3rd party has over BI/DS integration at the moment is the granular level that you can customize it to, but like I said the BI/DS integration and customization is getting better with every update.

Just think when BI/DS integration first came out, it was simply send a pic to DS for analysis, then came the option to send up to 5 pics at one second intervals, then came a revision to time the timing of the additional pics, then custom models, etc. It is getting there and with tuning can be a great tool.

The DS/BI integration is leaps and bounds better than it was when it first came out.

We tend to see on this forum the issues folks have with it, but with the right field of view, it is spot on. I only bring up the problematic view I am dealing with, but it works without fail on my other camera views.
I'll still give it a shot with the indoor cameras because...well...they do not have AI anyways so something helps out
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,450
Reaction score
47,588
Location
USA
I think indoors it will work fine.

The biggest problem it was having was at night with anything at a distance, but the custom dark model helps a lot.

I think some of the complaints early on was unrealistic expectations.
 

handinpalm

Getting comfortable
Joined
Sep 21, 2016
Messages
679
Reaction score
1,432
Location
Tampa Bay FL
I have a handful of 5 year old 5231's that I replaced with 5442, and upgraded 1, 5231 to the new firmware for IVS AI (210711). I have this upgraded 5231 in the same inside room as my 5442 with new firmware. I have been testing it against themselves for triggers and find that the old 5231 is superior to the 5442 IVS trigger reliability (sample of 1). Kind of surprised me, but I find the 5442 IVS firmware is not ready for prime time for the locations I have been using them. The 5442 IVS at night is not reliable, even though there is good contrast. Of course, the resolution of the 5442 is quite superior. I plan on loading the new firmware on another 5231 and trying on another location. May want to try the new 5231 firmware for IVS. Sorry for being a little off topic for DS.

IPC-HDW5231R-ZE IPC-HX5X3X-Rhea_MultiLang_PN_Stream3_V2.800.0000018.0.R.210707
 
Last edited:

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,690
Location
New Jersey
The other thing that helps DS is running it with a GPU rather than relying on the CPU version. Detection times are dramatically reduced which improves detection substantially, both day and night. A good NVidia card, with a cuda core count over 1000 produces the best results.

I just upgraded to the latest version of DS yesterday. While it's still kind of early to say definitely, the new version does seem to be a little better both day and night, but the jury is still out on that here.
 

Pentagano

Getting comfortable
Joined
Dec 11, 2020
Messages
576
Reaction score
269
Location
Uruguay
I've been using deepstack in docker desktop for 11 months, took 15 mins to set up, really easy tbh.
AI tool with the 2 deepstack ports I have configured. Very customizable.

I was interested to see how the BI deepstacks compares. Installed windows DS cpu and configured BI to use this. Disabled the other ports and AI.

CPU usage was much higher, even with mode set to medium and on 1 standard model also. Using sub streams also on an i7 7700 with 32GB memory.

Was quite disappointed as I thought I could shut down my docker desktop to save cpu and memory.
Unless I did something wrong the windows deepstack uses more cpu power than the DS in docker .
Processing time was about the same.
 
Joined
May 1, 2019
Messages
2,215
Reaction score
3,504
Location
Reno, NV
I've been using deepstack in docker desktop for 11 months, took 15 mins to set up, really easy tbh.
AI tool with the 2 deepstack ports I have configured. Very customizable.

I was interested to see how the BI deepstacks compares. Installed windows DS cpu and configured BI to use this. Disabled the other ports and AI.

CPU usage was much higher, even with mode set to medium and on 1 standard model also. Using sub streams also on an i7 7700 with 32GB memory.

Was quite disappointed as I thought I could shut down my docker desktop to save cpu and memory.
Unless I did something wrong the windows deepstack uses more cpu power than the DS in docker .
Processing time was about the same.
to be fair... the 3rd party program was also on the same Blue Iris windows machine, just to keep everything together. I was (still kinda am) unfamiliar with the do's/don'ts of linux/ubuntu so just wanted something to work as I was in a rush. DS worked for me quite well on the same windows machine as BI, except for rain/snowstorms where the CPU spiked quite often due to the constant DS analysis. That is what prompted me to swap to Dahua camera AI (and Wittaj's persistence :) )
With the Blue Iris updates that I see flying by rather often, I am sure BI will work it out with Deepstack eventually. Mike is not giving up on it ! That's a good sign.
 
Top