Hell Yeah! Direct Deepstack Integration - 5.4.0 - March 31, 2021

[cscoppa]

I saw that the cars going by are only hitting upwards of 44% from one of my cameras which is why they arent hitting. Weird because before this integration they would be hitting like 60%.

My backyard issue seems to be that it thinks my dog is a cat (she's a smaller dog, so not super surprising).

I get some funny ones too. We have a lot of Deer around here, Deepstack always flags them as "dog".

Also, this is me and my dog "bag of soil"

 

[Mike A.]

Also, this is me and my dog "bag of soil"

Good boy pat pat

 

[tech101]

I get some funny ones too. We have a lot of Deer around here, Deepstack always flags them as "dog".

Also, this is me and my dog "bag of soil"

View attachment 85948

Is there a certain setting for the labels to appear on playback ? I think i did try This check box that did not do it. Just wondering.

 

[cscoppa]

Is there a certain setting for the labels to appear on playback ? I think i did try View attachment 85949 This check box that did not do it. Just wondering.

I have that checked, but I don't see the box like my screenshot above unless I delete the video clip first, THEN go to that alert. Since it doesn't have the video it will show me that burned in alert image instead.

 

[Cameraguy]

Was having a couple issues but after a reboot suggested by @tech101 this is amazing. No cpu increase and deepstack is identifying objects great. Ken is a beast

 

[Vettester]

This check box that did not do it. Just wondering.

That’s the right option. Are you sure you have motion setup correctly?

 

[sebastiantombs]

I installed DeepStack this afternoon and was very pleased. Then things went down hill with the arrival of night time. It just doesn't seem to be able to identify anything in B&W. I have two cameras monitoring street traffic and it just doesn't identify a car at all. I've played with the "confidence" level and even shut down DeepStack and restarted it from PowerShell with high sensitivity. No dice. Any ideas? BU detects them with no problem while DeepStack just can't identify them in B&W with headlights or tail lights causing some bloom, apparently.

 

[wittaj]

@sebastiantombs - I had to tweak my nighttime BI settings. Added more make time and had it take an additional 5 images from the AI screen and that helped for the camera that would be looking straight down the driveway that would get side profiles of vehicles.

My other camera I tried it with that is positioned angled parallel to the road I ended up creating a clone and masking off in the distance as the headlight bloom would trigger from hundreds of feet away. But it too had trouble recognizing stuff. I had to slow the shutter and up the gain to make it more visible.

I think some angles will be problematic. But still a wonderful addition and another excuse to play and tweak in BI LOL!

 

[sebastiantombs]

Thanks @wittaj I spent an hour playing with camera settings and ran out of time. Both cameras are looking at fairly shallow angles at the road, say one at 20 degrees and the other at 40 degrees. Both get hammered with headlights and tail lights. The 40 degree one only has a short window to capture a vehicle, as in under three seconds, so increasing the number of images doesn't do much since they happen every second. I did try that as well.

I have one other camera that I'll clone and set up for vehicles. It's at right angles to the road but significantly further away so the size of a vehicle, or person, is kind of small. Trial and error at work with this one I guess.

 

[holiday]

installed this release 5.4.0.1, I was using "vorlonCD + AItool" prior to this.

Had to reinstall cpu version of deepstack for it to work.

Some cameras worked ok at first go but some didn't .. so have delete + new camera for it. check the log to see which camera needs work.

Seems be easy enough to get it working nicely.. thanks..

 

[Mike A.]

What's the basis for selecting CPU vs GPU?

I'm assuming that if you have a higher-end GPU that would be the way to go.

But I don't.

My BI server is an i7-6500 with a (somewhat pitiful) Radeon R5 340X (2MB). Card isn't used for anything. Server runs at about 14% baseline/20% max CPU depending on what's happening.

Stick with CPU or offload to GPU?

 

[D0T-C0M]

I tried deepstack tonight but to be honest I wasn't impressed, at least not so far. Noticeable decrease in motion sensitivity as compared with deeptrack off. I have a 5 second pre-trigger set and many videos start and I'm already well into the video with deepstack enabled. Also many false positives and false negatives. My garden hose was a person and my jeep was a suitcase, I know jeeps are kinda square but come on LOL. I will tinker with it some more and see if I can make it work. Seems like a lot of you here are having success with it maybe its just my settings.

 

[fenderman]

I tried deepstack tonight but to be honest I wasn't impressed, at least not so far. Noticeable decrease in motion sensitivity as compared with deeptrack off. I have a 5 second pre-trigger set and many videos start and I'm already well into the video with deepstack enabled. Also many false positives and false negatives. My garden hose was a person and my jeep was a suitcase, I know jeeps are kinda square but come on LOL. I will tinker with it some more and see if I can make it work. Seems like a lot of you here are having success with it maybe its just my settings.

Deepstack is used to fire alerts, not trigger motion detection. Motion detection should actually be set more sensitive when using deepstack. You need to use common sense when selecting what objects deepstack will be looking for. Not a suitcase in your driveway.

 

[D0T-C0M]

I did use common sense, I selected person,car,truck,bus.bicycle,cat,dog and deepstack detected my jeep wrangler as a suitcase and rejected the alert. I'm aware that deepstack shouldn't affect motion trigger sensitivity but when its on BI is noticeably less sensitive. I'll try increasing the sensitivity of the motion detection as you requested and see if it helps but its already at 312 and 24 which is pretty sensitive.

 

[austwhite]

I am a little bit of a newbie with Blue Iris, but I had been running the AI Tool with duplicate low res camera option previously.
I am finding that moving to the BI Integrated DeepStack seems to miss a lot that AI Tool picks up.
Does anyone have any (at least ball park) settings that detection could be set for the integration to work reliably, or is this just early stages and not quite yet up to production tasks?
I am using the same deepstack server in AI Tools as in the BI Integration, so I don't think this is a DeepStack issue, more either a setting or detection issue by BI.

Edit: I actually have two DeepStack servers running, one local on Windows (localhost) and one running in Docker on a VM. Both give same results.

 

[holiday]

keep your duplicate camera setting.
in the alert tab of your duplicate camera, under actions (for both on alert and on reset) trigger or cancel your main camera with the same http weblink used in the ai tool.

works for me this way.

 

[austwhite]

keep your duplicate camera setting.
in the alert tab of your duplicate camera, under actions (for both on alert and on reset) trigger or cancel your main camera with the same http weblink used in the ai tool.

works for me this way.

I tried this. I actually ran two identical cameras in the exact same location, same trigger settings (using Blue Iris motion trigger, not using ONVIF triggers) and found that BI Deepstack would miss somethings AI Tool Deepstack would trigger on, relevant things that is. It doesn't miss everything, but it seems to miss about 20% of items that I would want it to trigger on compared to AI Tool.
I might keep playing a little bit with this as it seems a lot of people are having very good success so it must be some setting on my end.

 

[fenderman]

I tried this. I actually ran two identical cameras in the exact same location, same trigger settings (using Blue Iris motion trigger, not using ONVIF triggers) and found that BI Deepstack would miss somethings AI Tool Deepstack would trigger on, relevant things that is. It doesn't miss everything, but it seems to miss about 20% of items that I would want it to trigger on compared to AI Tool.
I might keep playing a little bit with this as it seems a lot of people are having very good success so it must be some setting on my end.

What deep stack speed mode are you using with the AI tool?

 

[austwhite]

I did use common sense, I selected person,car,truck,bus.bicycle,cat,dog and deepstack detected my jeep wrangler as a suitcase and rejected the alert. I'll try increasing the sensitivity of the motion detection and see if it helps but its already at 312 and 24 which is pretty sensitive.

Increasing sensitivity of motion will not help DeepStack detection if it gets the object wrong. You could try running DeepStack in HIGH mode instead of Medium mode. It uses a little (not much) more CPU and is a touch slower processing, but is more accurate with its detection.
If links are allowed: Using DeepStack with Windows 10 (CPU and GPU) | DeepStack v1.2.1 documentation has the informatiion

 

[austwhite]

What deep stack speed mode are you using with the AI tool?

In both the Deepstack Docker and the Windows version of Deepstack, I have manually started up both with MODE=High option. I do not use AI Tools to autostart Deepstack, I have that starting separately outside of AI Tools to give a fair comparison.

Edit: Same with BI. I don't have BI starting deepstack so both AI Tools and BI Integration use same settings on deepstack