Ok. That makes sense. Thanks for the explanation. On Wednesday, December 18, 2013 10:19:43 AM UTC-5, Josh Warner wrote:

For completeness, also note earlier SLIC versions automatically forced a small Gaussian blur to the image prior to segmentation. This was controlled via the sigma parameter, and it defaulted to 1.

Now the default is sigma=0; i.e. SLIC just performs SLIC by default. This is more intuitive and offers compatibility with workflows including their own custom pre-processing blurs. So, in addition to what Juan noted above, if you want result parity with 0.8.x you also must set sigma=1.

On Wednesday, December 18, 2013 8:11:06 AM UTC-6, Juan Nunez-Iglesias wrote:

The segmentation did change, as a result of this PR:

...

https://github.com/scikit-image/scikit-image/pull/666

In summary, the handling of the ratio/compactness parameter in scikit-image's SLIC was different from the reference implementation provided by the authors, and we fixed that between 0.8 and 0.9. If you fiddle with the compactness parameter, you should be able to get something close to your original segmentation.

I hope this helps!

Juan. — Sent from Mailbox https://www.dropbox.com/mailbox for iPhone

On Thu, Dec 19, 2013 at 1:08 AM, scott.ri...@visionsystemsinc.com < scott.ri...@visionsystemsinc.com> wrote:

...

I recently upgraded skimage from release 0.8.2.0 to 0.9.3 and noticed that I am getting back a different segmentation than I used to.

I see that @jni and @ahojnnes have made quite a few edits to skimage/segmentation/slic_superpixels.py and _slic.pyx, so I suspect that is expected, but I wanted to make sure it wasn't a regression.

thanks Scott

-- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image...@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.