Regarding Vision Processing Interface SDK for python

Hi,

I tried to install vision programming interface from this page here (VPI - Vision Programming Interface: Installation)

I am unable to get the python bindings associated with the virtual environment, due to that I am unable to import the library for GPU based computations.

I would like to know what exactly are the requirements (eg: cuda version, type of gpu etc ) and if it works with any version of cuda or gpu type please help me in linking the python bindings with the virtual environment.

I installed the latest version as mentioned. libnvip2.3

Hi @vinay5, welcome back to the forums.

Did you see this post Missing VPI in Python (VPI 2 DP)

Maybe some of the hints will help there?

With respect to minimum versions I cannot say right away, I need to ask the experts.

Thanks.

Hi Markus,

Is this VPI package specifically for Jetson board?

Or will it work with other cpu’s as well which has nvidia gpu?

And also I’ll check the forum thread you asked me to check.

Thanks for the response

Hi again,

regarding requirements the minimum CUDA version is CUDA 11, recommended 11.4 and any GPU that supports this, which is any GPU after Maxwell, although at least Turing generation is recommended.

VPI was developed with Drive, Jetson and Edge computing platforms in mind, but it can still be used on x86_64 platforms like desktop CPU/GPU setups.

Hi,

Thankyou for ur response. I’ll check the folders specified once again and let you know ASAP.

Hi,

So now I am able to get the proper python bindings into my environment file.

But the problem I am facing is that it is not installed as a library from which I can import modules.

No module named ‘vpi.Image’; ‘vpi’ is not a package

This is the error I am getting.

I tried using Gaussian_blur functionality but couldn’t use it because vpi is not there as a package. Is there any pip install command from which I can install it?

Did you check the Quick Python Tutorial?

Try import vpi in the interactive Python shell and see if that works. If yes, then things should work.

If not, check the link to the other forum topic again please.

And I don’t think there is an API called vpi.Image. It should be vpi.asImage.

Hi,

That I sorted out, because python samples have different syntax.

But now other problem is I am unable to perform Gaussian_blur as it is not supporting the BGR image format, it says it can only work with Gray scale format.

blur = input.convert(vpi.Format.BGR8).gaussian_filter((0, 0), 2.0, border=vpi.Border.ZERO).cpu()
ValueError: VPI_ERROR_INVALID_IMAGE_FORMAT: Image format not supported by SeparableConvolution/CUDA

I need this to be working on BGR or RGB format as post this operation I have to use that array not single channel image.

Please let me know.