* check compiler support
* check HW support before executing
* add test doing round trip conversion from / to FP32
* treat array correctly if size is not multiple of 4
* add declaration to prevent warning
* make it possible to enable fp16 on 32bit ARM
* let the conversion possible on non-supported HW, too.
* add test using both HW and SW implementation
Enable "useProvidedKeypoints" flag on cuda::ORB's detectAndCompute function in order to able to describe provided keypoints.
* additional author : Ender Tunç <endertunc@yahoo.com>
resolves#6473
* Enable "useProvidedKeypoints" flag on cuda::ORB's detectAndCompute function in order to able to describe provided keypoints.
Update for using 'const' reference to avoid array copy on each iteration.
resolves#6473
Update for 'const' reference to avoid array copy on each iteration.
This is found in the original akaze repo. Previous sub pixel localization method assumes the coordinate (0, 0) is the up-left corner of the up-left pixel. But as far as I know, opencv uses the center of the up-left corner, in this case it should be done in this way.
35aeb83a71db3dc22981
Command line generation routine uses "-isystem" to include headers outside of "<opencv>/modules" folder, but GCC 6 does not work when passed "-isystem /usr/include" option.
Command line generation routine uses "-isystem" to include headers outside of "<opencv>/modules" folder, but GCC 6 does not work when passed "-isystem /usr/include" option.
All of these: (performance) Prefer prefix ++/-- operators for non-primitive types.
[modules/calib3d/src/fundam.cpp:1049] -> [modules/calib3d/src/fundam.cpp:1049]: (style) Same expression on both sides of '&&'.
Fix for inconsistent asserts in cv::fisheye::initUndistortRectifyMap() which prevents from passing empty matrices in debug build (which is allowed according to the code bellow the asserts and the docs).
