IP Cam Talk Custom Community DeepStack Model

[Tinman]

I HIGHLY recommend you delete your deepstack install and all its folders and then do a fresh install of deep stack once you get your card installed. It appears to me you have a corrupt install right now as you should not be seeing results like this with that processor. I did this awile back when they released a updated version of deepstack, it can't hurt.

 

[LopezEL]

I HIGHLY recommend you delete your deepstack install and all its folders and then do a fresh install of deep stack once you get your card installed. It appears to me you have a corrupt install right now as you should not be seeing results like this with that processor. I did this awile back when they released a updated version of deepstack, it can't hurt.

I agree. My deepstack analysis times are in the 100-200ms with an I5 10400 CPU. I am running the combined model.

When BI triggers and sends the pictures to deepstack for analysis, the CPU does increase close to 100% but it drops immediately.
I am only analyzing 3-4 real time pictures for trigger confirmation at 250ms intervals.
Using deepstack on 7 of 12 total cameras in my BI system. Have not had any issues with overloading the CPU. It's usually sitting at 7% utilization.

 

[sebastiantombs]

That's awesome. How many cameras do you have? And where did you get the night model? I only saw the three on the original post.

I have 21 cameras running right now. Probably 15, or more, are running DeepStack simultaneously. With the 970 the CPU still spikes, but only by an additional 5-20% depending on how many cameras trigger. Normal utilization on my system is around 20% without DS running analysis so say 25-40% with analysis.

The dark model is available from the DeepStack site or I'll zip it and PM it to you if you'd like to try it.

GitHub - OlafenwaMoses/DeepStack_ExDark: A DeepStack custom model for detecting common objects in dark/night images and videos.

A DeepStack custom model for detecting common objects in dark/night images and videos. - OlafenwaMoses/DeepStack_ExDark

github.com

 

[MikeLud1]

I have 21 cameras running right now. Probably 15, or more, are running DeepStack simultaneously. The dark model is available from the DeepStack site or I'll zip it and PM it to you if you'd like to try it.

GitHub - OlafenwaMoses/DeepStack_ExDark: A DeepStack custom model for detecting common objects in dark/night images and videos.

A DeepStack custom model for detecting common objects in dark/night images and videos. - OlafenwaMoses/DeepStack_ExDark

github.com

I am going to add the dark images to the custom models. Right now I am busy with the ALPR project.

 

[randylight]

I have 21 cameras running right now. Probably 15, or more, are running DeepStack simultaneously. With the 970 the CPU still spikes, but only by an additional 5-20% depending on how many cameras trigger. Normal utilization on my system is around 20% without DS running analysis so say 25-40% with analysis.

The dark model is available from the DeepStack site or I'll zip it and PM it to you if you'd like to try it.

GitHub - OlafenwaMoses/DeepStack_ExDark: A DeepStack custom model for detecting common objects in dark/night images and videos.

A DeepStack custom model for detecting common objects in dark/night images and videos. - OlafenwaMoses/DeepStack_ExDark

github.com

Wow that’s a ton. Are these all for your house or are you running a business?

 

[randylight]

I HIGHLY recommend you delete your deepstack install and all its folders and then do a fresh install of deep stack once you get your card installed. It appears to me you have a corrupt install right now as you should not be seeing results like this with that processor. I did this awile back when they released a updated version of deepstack, it can't hurt.

Do you have any tips on installing the GPU version? I heard it can be tricky.

 

[medzec]

I live in the PNW and judging by the public ring and next door posted videos, there is a badge all scofflaws wear. More of a uniform actually:
Backpack
Please don’t remove backpacks from the custom on your next spin up.
If you see anyone walking around more than a mile from the nearest school wearing a backpack, they are suspect.
if anyone starts down my walkway and DS sees backpack the outdoor conversation bot kicks in. (They don’t even flinch with motion lights etc)

Wow - taking all that out should make it a lot faster. It is fun to try out the toothbrush LOL, but if removing 2/3rds of the objects speeds it up, that would be wonderful.

I would almost like to see two custom models - one for vehicles/humans and one for animals.

Labels To Keep:
person
bicycle
car
motorcycle
bus
truck

Animal Labels:
bird
cat
dog
horse
sheep
cow
bear

 

[randylight]

The 2060 will make a night and day difference. I'm using an old GTX970 and getting ~50ms detection times with the combined model. Detection times at night are >250ms with combined and dark.

Here's a screenshot of the deepstack analysis. It's looking at an image every 250ms for 20 images so I guess it makes sense that I'm getting those long analysis times. Is this how it's supposed to work? My graphics card came in today so I'm trying to get that installed now.

 

[MikeLud1]

Here's a screenshot of the deepstack analysis. It's looking at an image every 250ms for 20 images so I guess it makes sense that I'm getting those long analysis times. Is this how it's supposed to work? My graphics card came in today so I'm trying to get that installed now.

View attachment 115485

That looks correct, once you get the GPU setup the detection times will look alot better.

 

[randylight]

That looks correct, once you get the GPU setup the detection times will look alot better.

Do you have any tips on getting the GPU version installed and set up? I've heard that it can be a little tricky. Was it as simple as the CPU install?

 

[wittaj]

Do you have any tips on getting the GPU version installed and set up? I've heard that it can be a little tricky. Was it as simple as the CPU install?

A few more steps. You can't simply install the GPU and run it.

From the Deepstack website:

Install DeepStack GPU
To install and use DeepStack GPU version on your Windows machine, follow the steps below

• Ensure your machine has an NVIDIA GPU
• Download and Install CUDA 10.1 from NVIDIA
• Download and Install cuDNN from NVIDIA

Once the above are installed, download and run DeepStack GPU version for windows via the link below.

>> Download GPU version for Windows

 

[sebastiantombs]

Wow that’s a ton. Are these all for your house or are you running a business?

Those are just for the house. Our "yard" is over an acre and the house sits back from the street about 150'. There are two tree mounted cameras that cover the area near the street. On the house there are four more plus another on the driveway gate. In the back there are three on the house plus another three on a shed covering the rear of the yard and where we park our cars. There are another two on both ends of the house covering the side approaches. Inside there's another three, one for a common area and two in the basement.

I want to add a few more as well. One for the package area at the front door and one, or two, to watch the inside of the shed where we store all the yard equipment. I should also add one for the smaller shed where the other three that monitor the back yard are located to watch the generator inside of it. Plus there is a blind spot in the basement that could use another camera.

The three basic rules of video surveillance cameras-

Rule #1 - Cameras multiply like rabbits.
Rule #2 - Cameras are more addictive than drugs.
Rule #3 - You never have enough cameras.

 

[randylight]

Those are just for the house. Our "yard" is over an acre and the house sits back from the street about 150'. There are two tree mounted cameras that cover the are near the street. On the house there are four more plus another on the driveway gate. In the back there are three on the house plus another three on a shed covering the rear of the yard and where we park our cars. There are another two on both ends of the house covering the side approaches. Inside there's another three, one for a common area and two in the basement.

I want to add a few more as well. One for the package area at the front door and one, or two, to watch the inside of the shed where we store all the yard equipment. I should also add one for the smaller shed where the other three that monitor the back yard are located to watch the generator inside of it. Plus there is a blind spot in the basement that could use another camera.

The three basic rules of video surveillance cameras-

Rule #1 - Cameras multiply like rabbits.
Rule #2 - Cameras are more addictive than drugs.
Rule #3 - You never have enough cameras.

Good lord that’s incredible. I already feel the itch and am currently mapping out areas for new cameras. Two definitely isn’t enough and can tell there will be many more in my near future.

I’ll install the GPU version tomorrow and report back on how it’s doing.

 

[sebastiantombs]

Yeah, I started with one about five years ago

 

[105437]

Those are just for the house. Our "yard" is over an acre and the house sits back from the street about 150'. There are two tree mounted cameras that cover the area near the street. On the house there are four more plus another on the driveway gate. In the back there are three on the house plus another three on a shed covering the rear of the yard and where we park our cars. There are another two on both ends of the house covering the side approaches. Inside there's another three, one for a common area and two in the basement.

I want to add a few more as well. One for the package area at the front door and one, or two, to watch the inside of the shed where we store all the yard equipment. I should also add one for the smaller shed where the other three that monitor the back yard are located to watch the generator inside of it. Plus there is a blind spot in the basement that could use another camera.

The three basic rules of video surveillance cameras-

Rule #1 - Cameras multiply like rabbits.
Rule #2 - Cameras are more addictive than drugs.
Rule #3 - You never have enough cameras.

Sounds almost exactly like my situation.

 

[randylight]

That looks correct, once you get the GPU setup the detection times will look alot better.

So all of a sudden my computer is working perfectly. I think it just needed restarted. I shut it down to install a new M.2 SSD and when I turned it back on my process times haven’t gone above 95ms. I probably don’t even need the GPU anymore but I’m going to wait and see if it reverts back to how it was before I rebooted.

 

[sebastiantombs]

I wonder if the drive you replaced had bad sectors or was going bad.

 

[randylight]

I wonder if the drive you replaced had bad sectors or was going bad.

I didn’t even change the drive out. I just installed a new one to clone my smaller boot drive. It was the first time I had shit it down since I got my cameras so maybe it just needed a fresh start? Idk but it works flawlessly now. Maybe I’ll install that night custom model. At least I have a GPU I can use in my computer now.

 

[nz1984]

I'm running deepstack in a docker container on a 2gb Jetson nano - using this custom model is nice as it seems to use slightly less ram than the standard one. However, I can't get Blue Iris to address it correctly. I don't need to specify a folder for the custom model in BI since BI isn't controlling the command line to run deepstack. However, I do try running with it checked because I assume that's what prompts BI to call http:// 192.DEE.PST.ACK:80/v1/vision/custom/modelname as opposed to the default http:// 192.DEE.PST.ACK:80/v1/vision/detection. Is there a way I can see what the command looks like that BI is sending out to deepstack? The deepstack log in BI isn't particularly helpful showing only $SysReset and Error -1:



Global AI Settings:



Camera settings:

 

[MikeLud1]

I'm running deepstack in a docker container on a 2gb Jetson nano - using this custom model is nice as it seems to use slightly less ram than the standard one. However, I can't get Blue Iris to address it correctly. I don't need to specify a folder for the custom model in BI since BI isn't controlling the command line to run deepstack. However, I do try running with it checked because I assume that's what prompts BI to call

http://192.DEE.PST.ACK:80

/v1/vision/custom/modelname as opposed to the default

http://192.DEE.PST.ACK:80/v1

/vision/detection. Is there a way I can see what the command looks like that BI is sending out to deepstack? The deepstack log in BI isn't particularly helpful showing only $SysReset and Error -1:

View attachment 115607

Global AI Settings:

View attachment 115608

Camera settings:
View attachment 115609

I am not sure on have to get the custom models to work on a Jetson, email Blue Iris Support and ask them. I do remember somebody having the same issue and they were able to help.