Refactoring code according to refactored inferenc-engine-interface
NOTE that OpenCV doesn't support to get information of input layer.
The input should be set by calling cv::dnn::Net::setInput() API.
OpenCV allocates memory on the first cv::dnn::Net::forward().
After that, it reuse the allocated memory.
So, in this refactoring,
Create mInputData as cv::Mat type for input tensors
and returns their allocated memories.
Allocate the memories of mOutputBlobs by calling forward() with
a dummy tensors.
Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>