Hello
Jetpack 5.1.2 comes with VPI and OpenCV without CUDA support that can be used with system Python.
Using system Python, we run a test script including simple OpenCV operations.
In the test script we import both cv2 and vpi.
We do not get any error even cv2 or vpi is imported first.
For our own virtual environment, we build OpenCV with CUDA support.
When we run the same script with after first importing cv2 then importing vpi we do not get any error.
But when we first import vpi then cv2 we get error.
While using vpi and cv2 together should we stick to cv2 non-cuda version?
Hi,
This is constraint in python. For more flexibility, may consider use C language.
Hi again,
The problem occurs when python environment has links for system vpi and custom OpenCV built with CUDA for version 4.8.1.
We built OpenCV with CUDA for version 4.8.1 since it is supported as we observe in Jetson Containers.
Then we built OpenCV with CUDA support for version 4.5.4, which is also the version that comes with Jetpack 5.1.2.
When we add system vpi link and link for OpenCV with CUDA support for version 4.5.4 into our environment we do not observe the error.
I think OpenCV version of the system(vpi) and custom OpenCV built must be same.