DeepStack Windows Version CPU and GPU Beta

john · December 19, 2020, 1:27pm

Hello everyone, I am excited to share today a major milestone, DeepStack Windows Edition for both CPU and GPU.

This release is very important to us and the community because it enables running DeepStack on Windows without using Docker.
Also, for the first time ever, you can now run DeepStack GPU on windows without using WSL.

While we update the docs to reflect instructions for this release, installation and usage instructions are available here https://github.com/johnolafenwa/DeepStack/releases/tag/0.1-beta

Thanks everyone for the patience.
We have worked hard to make this Beta very stable, however if you face any issues, please post here so we can make make the final version much more better.
Can’t wait to see what you all do with this.

mrpie · December 19, 2020, 3:01pm

Thank you very much for the update.

Do we need to change anything in the code for object detection? I read the face recognition docs and I didn’t note any changes in the code examples.

I’m asking because for some reason my Blue iris setup stopped working. When it detects something Deepstack receives the request but then something happens because the BI tool closes and I’m attaching what deepstack shows.

If I debug on visual studio it says:

I turned off the send_errors option, but I’m still getting this error.
I’m not an expert but I’m understanding that deepstack is returning a null string.

Looking at the log I can see that it gets stuck in step 2/6

Log("(2/6) Waiting for results");
var jsonString = await output.Content.ReadAsStringAsync();
Response response = JsonConvert.DeserializeObject(jsonString);
                        if (!response.success)
                        {
                            Log(response.message);
                            return;
                        }

john · December 19, 2020, 3:28pm

Hello @mrpie, the error above indicates the detection endpoint is not active.
If you are running the windows version, ensure you start DeepStack with --VISION-DETECTION True

mrpie · December 19, 2020, 3:35pm

Oh I see. So now the question is how do I start it with face and object recognition?
When I open my cmd and type

deepstack --VISION-FACE True --PORT 80
it doesn’t let me to enter another command. So I opened another cmd and typed

deepstack --VISION-DETECTION True --PORT 80
which gave me this, I think this is not the right way to do it.

And many thanks for your help.

MissMusic · December 19, 2020, 3:38pm

Try different ports? You can’t run two instances both on port 80.
EDIT:
Never mind, i could see you actually ran 81 on one from the screenshot.

john · December 19, 2020, 3:43pm

You can run both face and object detection as below

deepstack --VISION-DETECTION True --VISION-FACE True --PORT 80

mrpie · December 19, 2020, 3:59pm

Perfect! This is awesome, now it takes less than half the time to analyze
Now I will play with custom models.

john · December 19, 2020, 3:57pm

Great! Would love to know how the custom model works for you with this.
BTW, are you using cpu or gpu version?

mrpie · December 19, 2020, 3:59pm

I will let you know then!
I’m using CPU version

Tinbum · December 19, 2020, 6:26pm

Running the GPU version, times are fantastic about 75ms. Thank you.

Can’t wait for the Pi version.

MissMusic · December 19, 2020, 7:03pm

GPU version kicks ass! Average about 50ms, it even does full 3840 x 2160 images in less than 200ms.
Running on a P2000

Harby230 · December 19, 2020, 8:42pm

Excited to see the GPU version for windows, however I am having trouble.

Info:

Nvidia 1070
Windows 10 - 20H2
Cuda_10.1.105_418.96_win10
cudnn-10.1-windows10-x64-v8.0.5.39
Deepstack GPU 2020.12.beta

I can’t get scene detection to work. Object detection is working perfectly. Requests to Scene detection just hang - no response returns ever. I’ve tried this on two different machine (1070 on one, 2x1050 TI on the other, no other differences).

I don’t see anything in the logs out of the usual - just the startup log lines, then nothing while the scene requests come in. If I run object detection requests, they show up in the logs as expected. Have tried a variety of images, no success. The same images and requests work when sent to the old CPU docker instance.

Running with:
deepstack --VISION-SCENE True --VISION-DETECTION True --PORT 8086

mrpie · December 20, 2020, 1:33am

I was able to train a custom model using google colab, but when I enter the command to register the new model, it seems that I’m doing something wrong because it doesn’t show me the v1/vision/custom/mymodel as on the video. It only shows me v1/backup and v1/restore

Tried using different ports and sintax but got the same result. I’m pretty sure I’m close to be able to deploy it.

shallowstack · December 20, 2020, 3:30am

Do we need to install Visual Studio 2017 for the CuDNN CUDA 10.1 libraries to work properly?
https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-windows
section 3.3.5

My object detection seem to load the correct API (v1/vision/detection loads using deepstack --VISION-DETECTION True --PORT 80 --API-KEY myKEY), but just hangs and doesn’t respond to queries. Did i botch the install?

using

Nvidia Quadro FX1800
Nvidia Driver Version 342.00
Windows 10
Cuda_10.1.105_418.96_win10
cudnn-10.1-windows10-x64-v8.0.5.39
Deepstack GPU 2020.12.beta

Thanks

shallowstack · December 20, 2020, 12:05pm

If I have a chance I will swap out to a spare quadro K2000 (keplar) GPU and see if that helps… Anyone running on old hardware?

Is Keplar architecture the first GPU to be compatible with CUDA 10.1?

CUDA SDK 10.0/10.1/10.2 support for Compute Capability 3.0 - 7.5 (Kepler, Maxwell, Pascal, Volta, Turing) Last version with support for compute capability 3.x (Kepler).

***Update - tried with Quadro K2000, installed correct drivers, installed VS 2017, reinstalled everything else, but getting the same problem - no response from deepstack.

when I type the IP:PORT into web browser it says the deepstack server is activated, so it should be running right?

Do i need to do anything else, or just run the below from powershell?
deepstack --VISION-DETECTION True --PORT 80

LiveITA · December 20, 2020, 9:15am

Has anyone tested if VorlonCD AI tool https://github.com/VorlonCD/bi-aidetection works with the new version?

Tinbum · December 20, 2020, 10:56am

@LivelTA yes works great with GPU version, not tried CPU as running docker but you have to start with Powershell.

Tinbum · December 20, 2020, 11:33am

Is there a way to start the GPU version automatically on Windows start up.

john · December 20, 2020, 11:49am

Hello @Harby230 your run command is fine.
Please retrieve logs from C:\Users\username\AppData\Local\DeepStack\Logs

You will find the file stderr.txt, open that file and share the contents here. If anything went wrong, it will be there

@shallowstack Please retrieve the logs from your setup as shown above, and share here.

john · December 20, 2020, 11:47am

Hello @mrpie, your run command should be

deepstack --MODELSTORE-DETECTION "C:\Users\angel\Document\modelos" --PORT 81

Read the release notes here for how to run deepstack windows https://github.com/johnolafenwa/DeepStack/releases/tag/0.1-beta