CUDA Debug does not break on breakpoint

My application uses OpenCV to load and pre-process images. If I call cv::resize() before launching kernel, breakpoints inside the kernel will not work when CUDA Debug is launched. They work if the cv::resize() is commented out.
To isolate the problem, I used default CUDA ‘hello world’ program and just called cv::resize() at the beginning. Same effect. Commenting out the resize call again fixed the problem. The project download can be found here:
https://mega.co.nz/#!4ZhH3ZiL!uGHEXJp8sgGKIimAul6mT9SOY1NZ7F87kUgWaziVU-g
I would appreciate if someone could try it out and tell me if the issue is reproducible on other machines. Also if anyone knows how to work around this issue/the actual cause, please reply.
My setup:
NSIGHT 4.5.0 Release Candidate 1 2015/01/13
GTX 780 Driver Ver. 347.25

Hi makule,

I’m sorry to say your project is blocked by NVIDIA policy. We cannot fetch it. Would you mind present your email address then we could send you a FTP link to upload your project to us? Thanks.

Hello, thanks for quick reply. You can send the FTP link to ds376@sussex.ac.uk. Thank you.

Link received, file uploaded. Looking forwards to more updates.

Hi makulu,

Thanks for uploading the project. We have reproduced your problem on our side. It’s an Nsight issue. Before we addressing it in new edition, you could try this workaround: Call any CUDA runtime API before OpenCV code, e.g.cudsSetDevice(). Then you could get CUDA kernel BP hit in CUDA debugging. Sorry for the inconvenience caused to you.

Hi makulu,

Thank you again for the patience. We have addressed and fixed this issue. Please look forward to Nsight 4.6.