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

Very cool.

I like getting my wildlife captures too though. I haven't looked at the Deepstack stuff much since I knew that as it stood it would be a rabbit hole for me (no pun intended). lol

I'm assuming that you can set tags for wildlife in some way as well? Does it need to be trained for that?
 
  • Like
Reactions: sebastiantombs, 105437 and tech101
Anyone know why if I run deepstack I just get this window?

But if I run command prompt and paste in this:
deepstack --VISION-DETECTION True --PORT 83

It works as long as I keep command window open
 

Attachments

  • PXL_20210401_005555323.MP.jpg
    PXL_20210401_005555323.MP.jpg
    6.2 MB · Views: 118
Mike A. said:
Very cool.

I like getting my wildlife captures too though. I haven't looked at the Deepstack stuff much since I knew that as it stood it would be a rabbit hole for me (no pun intended). lol

I'm assuming that you can set tags for wildlife in some way as well? Does it need to be trained for that?
Click to expand...
I'm with you, I like to capture wildlife since I live in a rural area. Deepstack doesn't offer "labels" for common wildlife like deer, coyote, fox, skunks etc.
 
  • Like
Reactions: bluemystic
Cameraguy said:
Can we add titles to the blue iris dialog box?
Click to expand...
Yes, here's what Deepstack can identify.

person, bicycle, car, motorcycle, airplane,

bus, train, truck, boat, traffic light, fire hydrant, stop_sign,

parking meter, bench, bird, cat, dog, horse, sheep, cow, elephant,

bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase,

frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove,

skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork,

knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot,

hot dog, pizza, donot, cake, chair, couch, potted plant, bed, dining table,

toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave,

oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear,

hair dryer, toothbrush
 
  • Like
  • Wow
  • Love
Reactions: netcsk, Nate Scott, VorlonFrog and 3 others
105437 said:
Yes, here's what Deepstack can identify.

person, bicycle, car, motorcycle, airplane,

bus, train, truck, boat, traffic light, fire hydrant, stop_sign,

parking meter, bench, bird, cat, dog, horse, sheep, cow, elephant,

bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase,

frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove,

skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork,

knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot,

hot dog, pizza, donot, cake, chair, couch, potted plant, bed, dining table,

toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave,

oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear,

hair dryer, toothbrush
Click to expand...

How about "animals" to cover all.....well, animals?
 
Til I can test more I'm going back to my docker running deepstack and ai tool.. I will say the new bi integration saves a ton of cpu tho
 
  • Like
Reactions: sebastiantombs
OK I couldn't resist and gave it a spin as nighttime is where I am most interested in this working.

Personally I still find the AI in the camera to still be superior at the moment (which I expected as Deepstack itself hasn't changed, but would also like this for my non-AI cameras).

I too did not see a huge jump in CPU running a 4th generation, but certainly more than a newer machine would probably show.

I can confirm that as suspected, Deepstack does require images (as opposed to pulling from the video stream) and it is pulling the alert images. It maxes out at pulling 5 images at 1 second intervals for sending to Deepstack. So depending on your camera location setup, you may still miss AI events.

It isn't the absolute answer to AI and unlike in the Dahua cams with AI where you just check a box for human or vehicle, this Deepstack integration still requires a lot more tweaking within BI to get results.

I cloned two cameras and set them up with minimal motion settings just to see what happens. My camera that is facing straight out at the street to get side profiles of vehicles missed every vehicle as my house is in a tiny dip, so I get a lot of headlight shine on the road prior to the vehicle entering the image, and even the 5 second images missed them.

I had to set a longer make time and define minimum sizes to get it to work reliably. Not a deal killer, but certainly not as easy as checking a box like in a Dahua AI cam. For people new to BI, I think this will be a learning curve to get it to work reliably.

This straight on shot down the driveway once dialed in did a good job of identifying people or cars.

My next cam I tried is one angled looking parallel to the street to see vehicles coming and going. This is serving as a spotter cam for my PTZ and the camera AI is spot on for this one, so I was anxious to see if I could mimic it.

Deepstack failed miserably with this viewpoint. I got it picking up boats that didn't exist and missed people 50% of the time. Missed all the vehicles as the headlights would trigger the motion from hundreds of feet away.

Still working on trying to dial this one in. Will have to cut the area of motion detection way back, but too much and then BI misses the motion and thus so will Deepstack.

As with most of what we do, YMMV and I can certainly see image areas where this is a wonderful addition to an already great program! But I know there will also be camera locations where this will be problematic.

But glad to have this introduced to the program. Great feature!
 
Last edited:
  • Like
Reactions: anijet, gwminor48, sebastiantombs and 2 others
wittaj said:
OK I couldn't resist and gave it a spin as nighttime is where I am most interested in this working.

Personally I still find the AI in the camera to still be superior (which I expected as Deepstack itself hasn't changed, but would also like this for my non-AI cameras).

I too did not see a huge jump in CPU running a 4th generation, but certainly more than a newer machine would probably show.

I can confirm that as suspected, Deepstack does require images (as opposed to pulling from the video stream) and it is pulling the alert images. It maxes out at pulling 5 images at 1 second intervals for sending to Deepstack. So depending on your camera location setup, you may still miss AI events.

It isn't the absolute answer to AI and unlike in the Dahua cams with AI where you just check a box for human or vehicle, this Deepstack integration still requires a lot more tweaking within BI to get results.

I cloned two cameras and set them up with minimal motion settings just to see what happens. My camera that is facing straight out at the street to get side profiles of vehicles missed every vehicle as my house is in a tiny dip, so I get a lot of headlight shine on the road prior to the vehicle entering the image, and even the 5 second images missed them.

I had to set a longer make time and define minimum sizes to get it to work reliably. Not a deal killer, but certainly not as easy as checking a box like in a Dahua AI cam. For people new to BI, I think this will be a learning curve to get it to work reliably.

This straight on shot down the driveway once dialed in did a good job of identifying people or cars.

My next cam I tried is one angled looking parallel to the street to see vehicles coming and going. This is serving as a spotter cam for my PTZ and the camera AI is spot on for this one, so I was anxious to see if I could mimic it.

Deepstack failed miserably with this viewpoint. I got it picking up boats that didn't exist and missed people 50% of the time. Missed all the vehicles as the headlights would trigger the motion from hundreds of feet away.

Still working on trying to dial this one in. Will have to cut the area of motion detection way back, but too much and then BI misses the motion and thus so will Deepstack.

As with most of what we do, YMMV and I can certainly see image areas where this is a wonderful addition to an already great program! But I know there will also be camera locations where this will be problematic.

But glad to have this introduced to the program. Great feature!
Click to expand...
The camera is also not analyzing the video. Its not looking at every frame at 30fps. Deepstack has many options. For example, there is a high/medium/low speed mode option. The high mode is most accurate. By default it launches in medium. This is likely the mode BI is using. Eventually he will likely allow you to select the mode. If deepstack keeps up with the development it will only keep getting better. Also keep in mind that any area not covered by a BI zone is automatically masked and not analyzed.
 
  • Like
  • Love
Reactions: Flintstone61, JNDATHP, looney2ns and 5 others
Absolutely I believe that Deepstack and/or any add-ons like folks have created have come a long way in a short amount of time and with continued development will surpass what a camera AI is capable of. And they have created some cool tools and additional categories that I do not see a camera getting. We are already seeing people max out camera CPUs trying to run everything that a camera is rated for, so I do not see cameras getting more CPU power just for AI purposes, and certainly not to all the categories that Deepstack is performing!

Heck I think Deepstack is already better than the Sentry option when I tried a demo of it!

I am sure people have experienced opposite of what I have as well. Like anything, there are locations where camera AI probably sucks. I made sure to position mine for as optimal as possible.

But this is a great addition and certainly less CPU intensive than the prior methods of incorporating it into BI. And I will continue tweaking my settings because it is a great addition and amazed he was able to pull this off and doesn't seem buggy for a new addition! Awesome stuff!
 
Last edited:
  • Like
  • Love
Reactions: Flintstone61, JNDATHP, tech101 and 1 other person
Could we have an option to connect to external DeepStack using an ip:port please?
I have my BI in a VM without GPU (Direct write to disks using only 2 cpu cores!! ) and i have GPU dedicated to dockers, So i can have a deepstack docker utilizing GPU.

Also there is a type in Camera Settings
"Additioanl"
1617254952942.png
 
  • Like
Reactions: juliandom and BORIStheBLADE
Changed my setup last night to go purely with BI only. Everything is working great. The only capability that's missing compared to running AI tool that I can think of is only recording a clip when the AI has detected a selected object. Right now you'll still record a ton of clips if an a-hole spider decides to build a nest in front of your lens. Would be awesome if there was a checkbox on the Record tab to only record when Deepstack has confirmed an object.
 
  • Like
Reactions: juliand, t_andersen, Buxton and 2 others
This is a real game changer! I've used AI Tools for a long time now and find it reduces false alerts greatly and with this implemented into BI its just great. If anyone ever bitches about the prices of BI then they need to F**k right off!
Try finding an alternative for this price point that offers so much!! Mr Ken, take a BOW :clap:
 
  • Like
Reactions: Nanookofthenorth, JNDATHP, looney2ns and 5 others
I'm running it with the settings as they came in BI (with the simple addition of ,cat,dog because critters...). So far as I'm concerned, the integration is amazing, and also, amazingly simple. I set up every outdoor and garage cam with it (19 to be exact), and I'll be darned if I'm noticing much increase in CPU on my old I7-3770 box.
 
  • Like
Reactions: looney2ns, anijet, sebastiantombs and 1 other person
Is there a way to use more than one Deepstack server in alternating fashion? I know AItool supports this but I read the BI help file and didn't see anything about using different servers.
 
  • Like
Reactions: juliandom
Considering he just implemented Deepstack, I think it is phenomenal what he has done. Certainly another continued reason to use this program. Takes away the "slight" advantage some may say the newer AI NVRs have....

As far as running Deepstack in a VM or alternating fashion, I think this implementation and the very low CPU requirements will fit the needs of most people so I doubt we would see a VM option soon, but who knows.

If you have already configured your camera motion settings to avoid a lot of false triggers, this integration and ease of use is perfect.

There are people that wish BI was created to run in VM (although some have figured out how) or on a Mac or some other platform and he hasn't done so yet.

But if you want to be able to run Deepstack in a VM or alternating fashion, then I suggest you send in a request to him. If it can be done and there is enough interest, he has proven to go the extra mile.
 
Last edited:
  • Like
Reactions: tech101
You must log in or register to reply here.
Top Bottom