conversion from PIL image to OpenCV image fails

[Zak Stone]

Binary package hint: python-opencv

The following method of setting the data of an OpenCV image fails with version 1.0.0-6.1ubuntu1 of libcv1 and python-opencv:

image = cv.cvCreateImage(size, cv.IPL_DEPTH_8U, 3)
image.imageData = data # set some data

In particular, this breaks the conversion from PIL images to OpenCV images in adaptors.py. I believe this issue has been corrected in more recent versions of OpenCV -- many helpful details about the problem are available here:

http://www.depthfirstsearch.net/blog/2008/09/22/opencv-and-python/

Thanks,
Zak

[Oliver Gerlich (ogerlich)]

The bug is known upstream (at http://sourceforge.net/support/tracker.php?aid=1618474) and is fixed in upstream CVS (at http://opencvlibrary.cvs.sourceforge.net/opencvlibrary/opencv/interfaces/swig/python/imagedata.i?r1=1.3&r2=1.4).

The patch from CVS is attached below. It applies without problems, but apparently the _cv.cpp file which is generated by Swig is not regenerated automatically. Adding --with-swig to the "configure" flags (for example in debian/rules, at the end of "confflags = " line) is a workaround for this.

[Matthias Klose (doko)]

fixed in 2.1 (natty)