Now for stitching 6 images (1280*960) I spend 5 - 10 seconds, each run different time.
And when it tooks 10 secs I've got result.tif with size **158 bytes** (header only?).
LARGE_INTEGER frequency; // ticks per second
LARGE_INTEGER t1, t2; // ticks
double elapsedTime;
[...] QueryPerformanceCounter(&t1);
Stitcher stitcher = Stitcher::createDefault(false);
Stitcher::Status status = stitcher.stitch(imgs, rois, pano);
if (status != Stitcher::OK)
{
cout << "Can't stitch images, error code = " << status << endl;
return -1;
}
QueryPerformanceCounter(&t2);
elapsedTime = (t2.QuadPart - t1.QuadPart) * 1000.0 / frequency.QuadPart;
printf("%.6f ms",elapsedTime);
imwrite(result_name, pano); [...]
And I have a two questions.
(1) Why it (different computation time) happened
(2) Is it possible to improve the speed?
PS: Unfortunately, I can't simply use true flag in Stitcher::createDefault, because my roi size is small (170x960), and as I know this is the issue for GPU SURF functions. But may be it's possible to use TBB or GPU in other sipmle steps (like cv::detail::BestOf2NearestMatcher)?
PPS: I'm, already tried to play with registrationResol, it helps a little, but not much.
↧