Help me get a hit on this clip?

[ProTapper]

Hi Friends,

My DS tuning revealed the identification below. My Push notifications are set to a very low confidence of only 35%. And DS, while tuning identified these fellas as cats, dogs, even birds and persons several times, however, ended up canceling the event/notification by ultimately saying, Nothing found! Do you guys see this on your ends?

Do you have tips on how I can catch events like this in the future?

Thank you.

 

[sebastiantombs]

When you figure it out be sure to let us all know. We're all looking for that answer.

 

[ProTapper]

When you figure it out be sure to let us all know. We're all looking for that answer.

Haha really? What did I miss? I thought I was just unlucky or not competent enough to fine tune DS and BI settings to get it... What keeps us from detecting this? Do you know? Even as cats? LOL!

 

[sebastiantombs]

There are a number of DeepStack threads, already, discussing this and other problems. Have a look around. If I knew how to fix it, the answer would already be available to everyone or I'd be working with Ken at Blue Iris.

I will say that from my experience with DS, so far, it has problems with lighting and contrast. Another thing seems to be related to exactly when the frames are captured and when the trigger actually happens in relationship to the motion causing the trigger. Check the other threads.

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

5.4.0 - March 31, 2021 Support for DeepStack integration (without the separate AI Tools package) via new Settings AI page. DeepStack may be used for object classification to confirm alerts containing persons, vehicles, pets, and more. Plate Recognizer ALPR configuration has been moved to the new...

ipcamtalk.com

5.4.9 - June 25, 2021

5.4.9 - June 25, 2021 Enhanced synchronization between alert images and alert processing. Alert images are not created until timer conditions are met on the Alert tab in camera settings. Processing of “re- triggers” is improved following a cancelled alert. Improved handing of DeepStack static...

ipcamtalk.com

AI/Deep Stack Inquiry

Ok, I have enjoyed my learning experience with Blue Iris. I think this is an awesome system with so many features I feel as though I will never master it. One thing I have wondered about is integration with Deep Stack. I get a lot of false alarms on my cameras (insects or birds a lot of times)...

ipcamtalk.com

5.4.8 - June 16, 2021 - A DeepStack status page has been added

5.4.8 - June 16, 2021 A DeepStack status page has been added. This allows you to inspect DeepStack results and performance following an alert confirmation or when using the testing & tuning options in the clip viewer. You may now specify 2 objects together with “and” logic on the “to confirm”...

ipcamtalk.com

 

[wittaj]

Or reread the DMs we have been having on the subject

 

[ProTapper]

Thanks @sebastiantombs and @wittaj (always breaking my legs!)... AJ - You and I are DMing about the inconsistency between the cameras. This, I thought was especially strange since it detects them as cats and I have cats as a check . But yes, as AJ would school me, the tuning and live processing have no actual relationship apparently.

You guys are onto something with the light and contrast thing. Also, definitely the timing of when the stills are sent to BI for processing. Both of these things, I understand via reading the threads and talking to AJ, are not necessarily controllable via BI yet?

Question is... Even if Ken enables the finer control via BI on these... Does this level of processing, specifically around lighting and contrast even exist in DS yet? Or is it just a matter of maturing those models under different lights over time?

Haven't seen DS getting many updates either...

Thanks as always, guys... Was worth a shot haha!

 

[wittaj]

Just giving you a hard time LOL.

The better defined the image with proper level of contrast, etc. (that we all are still working on LOL) will improve the DS process. And more light LOL.

And I am testing the new 5.4.9.2 and I am getting the blue border around the entire image showing which image is sent to DS. That is telling.

I now have (knock on wood) the other issues I have experienced with 5.4.7.X and above no longer happening!

Facial recognition still doesn't like custom model running too though.

EDIT - nope experiencing the computer issues I had when going past 5.4.6.3

 

[fenderman]

Hi Friends,

My DS tuning revealed the identification below. My Push notifications are set to a very low confidence of only 35%. And DS, while tuning identified these fellas as cats, dogs, even birds and persons several times, however, ended up canceling the event/notification by ultimately saying, Nothing found! Do you guys see this on your ends?

Do you have tips on how I can catch events like this in the future?

Thank you.

The potted plan in both images and the "cat" in the first image are stationary and will not trigger an alert unless you uncheck detect/ignore static objects. Its difficult to tell what its seeing as the bird.
Remember
"DeepStack objects recognized only when substantially overlapping areas where Blue Iris is
also tracking objects in order to reduce the detection of static objects."

 

[ProTapper]

Just giving you a hard time LOL.

The better defined the image with proper level of contrast, etc. (that we all are still working on LOL) will improve the DS process. And more light LOL.

And I am testing the new 5.4.9.2 and I am getting the blue border around the entire image showing which image is sent to DS. That is telling.

I now have (knock on wood) the other issues I have experienced with 5.4.7.X and above no longer happening!

Facial recognition still doesn't like custom model running too though.

EDIT - nope experiencing the computer issues I had when going past 5.4.6.3

Thanks AJ! Sorry to hear that the computer or the DS crashing issues are back, but the blue border is progress! There is a good bit of light on this side... As you recall, you always tell me how much you like the light here haha! Do you think the faster shutter could help? I'm also wondering if the object size from BI could be something to play with for these little guys?

The potted plan in both images and the "cat" in the first image are stationary and will not trigger an alert unless you uncheck detect/ignore static objects. Its difficult to tell what its seeing as the bird.
Remember
"DeepStack objects recognized only when substantially overlapping areas where Blue Iris is
also tracking objects in order to reduce the detection of static objects."

Thanks fenderman! I have the Camera Motion detection turned off to reduce over lap. And I thought DS just analyzes the objects with the area defined by the BI motion zones, no?

The Potted plants, yeah are somewhat stationary minus the wind... But the "cats" are just the raccoons moving across the driveway LOL! When you do a DS tuning you'll notice the boxes it paints around are somewhat delayed. So you may think the object was stationary, but I assure you, these pack of racoons were moving and upto no good . I can try to share the video clip with you guys too if y'all think it'll help?

Thanks again all of you!

 

[wittaj]

While you have good light, it also casts a long shadow and will cause DS to mis-identify objects at night. Most have experienced that at night to some degree. You need to pick one item you reliably want alerted to. For me it would be a person not a potted plant

In tuning, DS will draw rectangles around everything it detects. I went out with a toothbrush to see it find it in tuning LOL.

Some people have experienced issues using camera triggers and DS. It should be one or the other. Or a large field of view.

 

[ProTapper]

Thanks AJ! Can we get off the Potted Plant guys? LOL! That's just DS tuning at work . Is it that unreasonable to want person, car, cat dog detection? Perhaps at this time it is !? I do only have BI triggers, no Camera triggers coming in for motion.

Don't get me wrong guys... Every day I'm grateful for BI and DS. I've come a long way from the crappy Arlos thanks to people like @wittaj, @looney2ns and @sebastiantombs... But came here to see if there were any fresh ideas to keep getting better and better haha!

 

[wittaj]

Train your own model to your field of view!

 

[ProTapper]

Train your own model to your field of view!

Teach me, Sensei ?

 

[wittaj]

Well we first need a stable version above 5.4.6.3 LOL.

 

[ProTapper]

Well we first need a stable version above 5.4.6.3 LOL.

Well played Sir... LOL! Custom model, duh!

 

[sebastiantombs]

I adjusted the angle of view of one of my cameras yesterday so that is gets a slightly larger area of detection where I am most interested in. DS did detect a lot more successfully last night. Based on that preliminary result, I'd also say "screen time" has a definite effect of successful detection, especially under more difficult conditions. To be fair, I also swapped out that camera last weekend from a 2231R-ZS to a 5442T-ZE. There was initially no dramatic improvement in detection until I changed that view angle just a couple of degrees.

 

[ProTapper]

Thanks @sebastiantombs! Not sure if it's a placebo or not... But changing the additional pictures sent in by BI to DS to a 2 second interval might have helped me too...

 

[CamCrazy]

Train your own model to your field of view!

I would second this advice, did the same for several of my cameras and it works well. Also appears to be less strain on processing for the CPU since model is smaller I presume with less images trained. Link below:

Deepstack custom models