can opencv compiled with GPU support, fall back to CPU

[pks916]

Is there any way to make opencv use cpu if gpu not detected.
right now i tried it and it gives the DLL no found error

ImportError: DLL load failed: The specified module could not be found.?

i do have the access to cudnn and cuda toolkit , appropriate version

[cudawarped]

I'm not sure I understand the question, if the dll isn't loading you cannot access either the CPU or GPU functions?

ImportError: DLL load failed: The specified module could not be found.?

@pks916 Do you know why you are getting this error? Is it because the cuDNN and CUDA toolkit shared libraries (dll's) are not present on your system and/or can't be found? How did you build the openCV python bindings?

Note: If you don't know how to check see https://cudawarped.github.io/opencv-experiments/nbs/ImportError_dll_load_failed_while_importing_cv2.html

[pks916]

no gpu in the system
i put the dll files just for the path check in the config file so that it fails and switches to CPU
but instead of switching it exits and gives error mentioned above

[cudawarped]

The error does not relate to whether you have a GPU or not, it relates to whether the required shared libraries can be found. If you are seeing it check the link above to see which dll's are missing.

Are you using the pip install package and if not why not?

[pks916]

i wanted to use opencv with gpu and use dnn package so i am using opencv-cuda wheels
and i wanted it to check if gpu present then use cuda as backend otherwise cpu backend
it should work, right?

[cudawarped]

Yes but when you are unable to import OpenCV, see the link above to find out why.

[pks916]

so we can actually use model on cpu as well using the opencv with cuda, right?
when cv.cuda.getCudaEnabledDeviceCount() gives 0

[cudawarped]

You can run models on the CPU or GPU using the flags given in the documentation of the DNN module https://docs.opencv.org/4.x/d6/d0f/group__dnn.html.

[pks916]

but it is working fine when gpu is present and detected, that is the main point.
my other pc does not have graphic card where i want to run these models, but it gives error instead of running models on CPU.

i have the models initialise as gpu backend if cv.cuda.getCudaEnabledDeviceCount() else run on opencv backend

[cudawarped]

my other pc does not have graphic card where i want to run these models, but it gives error instead of running models on CPU.

What is the error? If its

ImportError: DLL load failed: The specified module could not be found.?

this has nothing to do with the GPU being present it is because the shared libraries can't be found see the link in my first reply to address this error.

[pks916]

ok i will see, but error is that only