It may get -nan value while calculate the feature of lsvm (Patch #1453)

Added by qunij - over 13 years ago. Updated almost 13 years ago.

Status:Done Start date:
Priority:High Due date:
Assignee:Vadim Pisarevsky % Done:

0%
Category:objdetect
Target version:2.4.0
Affected version: Operating System:
Difficulty: HW Platform:
Pull request:

Description

While the image have a block of the same value, the function of "normalizeAndTruncate(*,*)"(in file featurepyramid.cpp) will happen that zero divides zero(-nan).

This is excluded in the author's matlab code by adding small number(eps = 0.0001) before dividing.

The function "PCAFeatureMaps(*) may also have the same problem.

000860.jpg - This is the image that causes the issue. (75.6 kB) qunij -, 2011-10-28 03:41 pm

Associated revisions

Revision afd9523f
Added by Vadim Pisarevsky almost 13 years ago

possibly fixed bug #1453

History

#1
Updated by Alexander Shishkov almost 13 years ago

  • Tracker changed from Bug to Patch
  • Target version deleted ()

#2
Updated by Alexander Shishkov almost 13 years ago

  • Assignee deleted (Vadim Pisarevsky)

#3
Updated by Vadim Pisarevsky almost 13 years ago

without a test sample, it would take some extra time to reproduce the bug. However, I followed your advice and put extra "+FLT_EPSILON" in normalizeAndTruncate (SVN trunk, r7631). Looks like PCAFeatureMaps() should not have such problem.

If the bug is still there, please, reopen the ticket.

  • Status changed from Open to Done
  • Assignee set to Vadim Pisarevsky

#4
Updated by Alexander Shishkov almost 13 years ago

  • Target version set to 2.4.0

Also available in: Atom PDF