Looks interesting, I´m working on the same issue.
So, I can´t figure out what makes the difference. If I use:
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
The script break as seen on Abraham’s code (Assertion failed).
If those lines are ommited, the script goes on.
i very much doubt, that yours is the same…
how so, exactly ?? stacktrace ?
More explicit, we are working together. 
The goal is to be capables of doing image processing using opencv with GPU support from almost zero experience in opencv environment.
By now we´ve done several image processing using opencv cpu environment, but we cannot get the GPU environment to work, as we need “real time” FPS processing.
We have a Nvidia Tesla K20X but we can´t get the script to run. We are using multiple examples found by internet, and this may be the reason of the different codes we have.
Just now I cannot run any code because opencv is compiling, but the code that trows the error in my case is what you found in Adrian´s blog, just here:
Edit: We´ve been using Yolo v3 for object detection
Thanks for your time
sorry, it does not work like this here. we won’t read blog posts to find out about your error.
again, stacktrace (and exact code snippet), or get ignored.