Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.

EigenValsVecs question

Ruslan_G_
Beginner
669 Views

 

Hello! I compared results of EigenValsVecs functions in IPP and OpenCV and found that in some cases ippiEigenValsVecs function returns exactly (1.0f, 0.0f) and (0.0f, 1.0f) eigen vectors although OpenCV returns eigen vectors which are not even close to vectors obtained with IPP.

For example with input 5x5 matrix

-116.127525f,      -152.685074f,       108.699646f,     -168.813187f,      166.688339f,
-195.550934f,       142.426865f,         63.002094f,       67.792168f,        66.547188f,
   40.681679f,        91.366081f,        -10.763326f,      186.464005f,      164.351349f,
   44.740143f,        64.384964f,        -80.919853f,       -29.437082f,       45.215580f,
   95.272400f,        21.906961f,         84.793060f,       -50.961384f,      250.437256f

IPP returns (1.0f, 0.0f) and (0.0f, 1.0f) eigen vectors for the (3, 3) element of the matrix, but OpenCV returns (-0.540550, 0.841312) and (-0.841312, -0.540550).

Is there any reason for such behavior or some criteria when it appear?

0 Kudos
3 Replies
Chao_Y_Intel
Moderator
669 Views

Hi, 

Thanks for your report.  We are checking now.   Which version of IPP are you testing now? 

Regards,
Chao

 

0 Kudos
Ruslan_G_
Beginner
669 Views

Thanks for your answer. I'm testing IPP 9.0.

0 Kudos
Chao_Y_Intel
Moderator
669 Views

Rusian, 

thanks for the information.  The engineer is checking the code now, and I will provide you further feedback. 

Regards,
Chao

0 Kudos
Reply