cv::cuda::CascadeClassifier::create() runs very slow

edit

Hi everyone,

I'm working on porting my program from running on CPU to GPU. It runs on a NVIDIA Jetson TK1 board (Linux 14.04) with OpenCV 3.0.0. I'm using Cascade Classifiers to detect faces/body parts.

However I found that my program runs much slower since cv::cuda::CascadeClassifier::create() needs around 1.8s to process while its CPU counterpart only uses 0.1s. Is this normal?

Thanks!