DeepStack Case Study: Performance from CPU to GPU version

jeremypsl said:
What are those packages ?
Click to expand...
They are Python modules that DeepStack uses

These are optional, DeepStack will work without updating the 3 packages below
numpy-1.21.2-cp37-cp37m-win_amd64
Pillow-8.3.2-cp37-cp37m-win_amd64
scipy-1.7.1-cp37-cp37m-win_amd64

These are required to have DeepStack work with RTX3000 series GPUs. This link has more details what they are. PyTorch link
torch-1.9.1+cu111-cp37-cp37m-win_amd64
torchvision-0.10.1+cu111-cp37-cp37m-win_amd64
 
jeremypsl said:
Do i need to install Python from python.org or a package manager like anaconda to update those packages ?
Click to expand...
No, what needs to be done are in the post, link below.

You can skip deleting the highlighted folders
1638360461146.png

Deepstack GPU for Windows problems (Cuda version 11 and Cudnn install).

I have a 3090 RTX video card. I need a newer version 11 of CUDA because version 10.1 does not support a 3090 RTX video card. I am not sure if Deepstack is compatible with that version. Or that CUDA is even backward compatible. I also don't understand how I should install Cudnn (what folder...
ipcamtalk.com ipcamtalk.com
 
Gah!

1638410705645.png
 
What does everyone think about if we making a community DeepStack model. If we develop a list of objects that we want DeepStack to identify then everyone shares images with the object in them and we make one model that everyone can use.
Sample of the list:
person
car
truck
deer
raccoon
fox
opossum
cat
dog
coyote
 
  • Like
Reactions: 105437, prsmith777, TBurt and 2 others
sebastiantombs said:
Step 1 - install Nvidia CUDA capable card, preferably one with a large number of CUDA cores.
Step 2 - Follow the installation instructions for the GPU version as posted on the DS forum/page. You can skip the last step in those instructions regarding Visual Studio.
Step 3 - You're good to go.
Click to expand...

The GTX 1050 ti came yesterday and I threw it in my Dell Optiplex 7050 today.
It figures I'd botch the install as well. As best I could I followed the links provided by @jaydeel.
CUDA 10 failed repeatedly but CUDA 11 installed no problem after updating the GPU drivers from windows 10.
Installed zlib
Installed cuDNN
Copied the bin, lib and .h files to the appropriate directories.
Skipped the Visual Studio instructions..

Unfortunately, DeepStack is now only timing out after motion is detected.
Log tells me that Intel video hardware is detected and Nvidia hardware is detected.
 
AP-123 said:
The GTX 1050 ti came yesterday and I threw it in my Dell Optiplex 7050 today.
It figures I'd botch the install as well. As best I could I followed the links provided by @jaydeel.
CUDA 10 failed repeatedly but CUDA 11 installed no problem after updating the GPU drivers from windows 10.
Installed zlib
Installed cuDNN
Copied the bin, lib and .h files to the appropriate directories.
Skipped the Visual Studio instructions..

Unfortunately, DeepStack is now only timing out after motion is detected.
Log tells me that Intel video hardware is detected and Nvidia hardware is detected.
Click to expand...
Did you install the below
Download GPU version for Windows

1638751646394.png
 
MikeLud1 said:
What does everyone think about if we making a community DeepStack model. If we develop a list of objects that we want DeepStack to identify then everyone shares images with the object in them and we make one model that everyone can use.
Sample of the list:
person
car
truck
deer
raccoon
fox
opossum
cat
dog
coyote
Click to expand...
This would be great. I live in a rural area and would like DS to capture the wildlife that comes around our house. We have all of these animals plus skunks.
 
AP-123 said:
Yes, that's the version and instructions I used. I had to stop Blue Iris service for the install to complete.
Anything else I might try before I retrace my steps and install everything again?
Click to expand...
Since you installed Cuda 11 try updating the below Windows Packages. Steps are in the post that is linked below.

Manually update to the below Windows Packages
numpy-1.21.2-cp37-cp37m-win_amd64
Pillow-8.3.2-cp37-cp37m-win_amd64
scipy-1.7.1-cp37-cp37m-win_amd64
torch-1.9.1+cu111-cp37-cp37m-win_amd64
torchvision-0.10.1+cu111-cp37-cp37m-win_amd64

Deepstack GPU for Windows problems (Cuda version 11 and Cudnn install).
 
  • Like
Reactions: Paul_V and JNDATHP
Thanks Mike - I followed the instructions now have another error. HW VA not compatible and Deepstack responding at 127.0.0.1 looks off to me. My machine is on a 192.168.1.x ip.

1638768123831.png
 
@MikeLud1 - I saw that recent addition about instances, but I have left mine at 1.

I am running multiple cameras with DeepStack and works fine.

Are you saying if we change this number (and then go in to each camera and override the Server and change the port number) that it would result in faster responses?
 
wittaj said:
@MikeLud1 - I saw that recent addition about instances, but I have left mine at 1.

I am running multiple cameras with DeepStack and works fine.

Are you saying if we change this number (and then go in to each camera and override the Server and change the port number) that it would result in faster responses?
Click to expand...
I tried using two instances on my set up and I did not see much of a difference at all with response time. I'm sure my results are not typical since I'm running an older i5 with a P400 GPU.
 
wittaj said:
@MikeLud1 - I saw that recent addition about instances, but I have left mine at 1.

I am running multiple cameras with DeepStack and works fine.

Are you saying if we change this number (and then go in to each camera and override the Server and change the port number) that it would result in faster responses?
Click to expand...

I only found that using more DS instances consumed more of my PC memory with no benefits.
I do run 2 - but 1 is for a custom model and only active sunset-sunrise.
The other has all of my 8 cameras.

btw I use AI tool and deepstack GPU on a gtx970. Installed in 15 minutes and worked first time luckily.

But very interested to see how the 1050 card works.
Maybe try the AI tool and run the deepstack instances from there? It runs the GPU instances on startup and has more granular control.
 
You must log in or register to reply here.
Top Bottom