Updated by Alexander Shishkov about 13 years ago
Hello,
I use the Opencv-2.3.1a library, compiled with SSE, SSE2, SSSE3, cuda and qt, Ubuntu10.04 and have a intel quad core Q9650.
I used the BruteForceMatcher<L2<float> [[BruteForceMatcher]]<L2<float> > and its method match.
I observed that for several requests where to identical vectors were in the query and training set, that they weren't matched with distance 0. I tried some more examples, using only one vector for query and training (the same one), and achieved distances form -nan to 0 and some small >0 like 0.000247 (for example).
Then I tried the same binary from my code on several other machines also having Opencv-2.3.1 (I don't know which configurations were used) and it worked out fine there.
Is there an other dependent library on my machine broken?
If I use BruteForceMatcher<L2<double> [[BruteForceMatcher]]<L2<double> > all is fine, also for <L1<float> > there seems to be no problem.
I use the Opencv-2.3.1a library, compiled with SSE, SSE2, SSSE3, cuda and qt, Ubuntu10.04 and have a intel quad core Q9650.
I used the BruteForceMatcher<L2<float> [[BruteForceMatcher]]<L2<float> > and its method match.
I observed that for several requests where to identical vectors were in the query and training set, that they weren't matched with distance 0. I tried some more examples, using only one vector for query and training (the same one), and achieved distances form -nan to 0 and some small >0 like 0.000247 (for example).
Then I tried the same binary from my code on several other machines also having Opencv-2.3.1 (I don't know which configurations were used) and it worked out fine there.
Is there an other dependent library on my machine broken?
If I use BruteForceMatcher<L2<double> [[BruteForceMatcher]]<L2<double> > all is fine, also for <L1<float> > there seems to be no problem.