jakobwilm's profile - activity

As we know, OpenCV 3.0RC1 and 2.4.X are not API compatible. Hence, it would be very smart to use a new directory for the headers (e.g. opencv3/core.hpp). The same applies for the opencv.pc for package-config. It leads to all kinds of trouble when installing multiple versions of OpenCV. So what is the argument for not doing this?

Regards, Jakob

Hi Adam, in these measurements, how did you ensure synchronisation between GPU and CPU? The last plot would be consistent with the hypothesis that there is no sync, and hence part of the operations are measured in your copy-to-host timings. Regards, Jakob

Hello Community,

I am implementing a tracking application which extracts SIFT features from a calibrated stereo pair and matches them.

Now, I want the matching to obey the epipolar constraint, and also to increase robustness, I want only "bijective" matches -- i.e. use the crosscheck flag in cv::BFMatcher. Unfortunately, cv::batchDistance() throws an exeption because it does not allow crosscheck and a mask at the same time:

if( crosscheck )
{
    CV_Assert( K == 1 && update == 0 && mask.empty() );

Why can't batchDistance() run with swapped inputs and a transposed mask matrix? I currently have to do this myself:

matcher->match(descriptors1, descriptors0, matches10, mask.t());
matcher->match(descriptors0, descriptors1, matches01, mask);
cvtools::matchingCrossCheck(matches01, matches10, matches);

But I rather didn't because I can't implement this optimally without access to the underlying structures.

Can I post this as a feature request, or is there an alternative way?

Regards, Jakob Technical University of Denmark