Curtis - Thanks so much for digging. This focuses my efforts greatly. 

Tobias - If you see this email... I noticed that you are listed as author of 
the ConnectedComponents class. Any thoughts as I dig in?

Thanks,

Jay

> On Jun 30, 2015, at 12:34 PM, Curtis Rueden <[email protected]> wrote:
> 
> Hi Jay,
> 
> Interesting bug. I did some digging. The two Img<UnsignedByteType> instances 
> are identical in terms of dimensionality and pixel values, so that isn't the 
> culprit.
> 
> I also tried disabling the labeling computation for SCIFIO to see if it was 
> an issue of tainted state, but the order of computation doesn't seem to make 
> a difference either.
> 
> I also stepped a bit with the debugger and certainly the issue is inside of 
> ConnectedComponents.labelAllConnectedComponents. On line 115, the value of 
> numLabels comes back as 2 for the SCIFIO image and 1 for the IJ1 image. 
> Digging deeper becomes tricky due to the multithreadedness of the algorithm, 
> but I fear that's what you'll have to do in order to isolate the difference 
> in behavior.
> 
> Regards,
> Curtis
> 
> On Tue, Jun 30, 2015 at 1:01 AM, Jay Warrick <[email protected] 
> <mailto:[email protected]>> wrote:
> Hi All,
> 
> I am getting different behavior using the new ConnectedComponents class of 
> imglib2-algorithm depending on whether I get and Img<UnsignedByteType> from a 
> SCIFIO ImageOpener vs. when I get the same image as an Img<UnsignedByteType> 
> by wrapping an ImagePlus. In the SCIFIO case, ConnectedComponents finds the 
> expected single circle while with a wrapped ImagePlus, it doesn't find any 
> regions.
> 
> Any thoughts/fixes? Might there be something occurring during the wrapping 
> process to cause an issue? Here is a tiny commented maven project with the 
> example scenario.
> 
> https://github.com/jaywarrick/ConnectedComponentsTest 
> <https://github.com/jaywarrick/ConnectedComponentsTest>
> 
> The tiny .tif I'm using is included in the project but is loaded by a string 
> path. Thus, you need to retype the path to the image to match your computer, 
> but that should be it. Sorry... wasn't sure how to avoid that easily.
> 
> Thanks for your help,
> 
> Jay
> 
> 
> _______________________________________________
> ImageJ-devel mailing list
> [email protected] <mailto:[email protected]>
> http://imagej.net/mailman/listinfo/imagej-devel 
> <http://imagej.net/mailman/listinfo/imagej-devel>
> 

_______________________________________________
ImageJ-devel mailing list
[email protected]
http://imagej.net/mailman/listinfo/imagej-devel

Reply via email to