Now the output has changed but it is still not a baseline image. The images attached show the new output, some sort of a thresholded image that was probably used in mask computation. For reference the actual baseline (from when tensors were computed) is the second screenshot.

Fixed in r24673
See http://viewvc.slicer.org/viewvc.cgi/Slicer4?view=revision&revision=24673

Thanks to "Orjan Bergmann", the issue is now fixed and the outputs (baseline and mask) of the modules are now tested for regression.

Bonus: It would be great to add an other test checking if the "removeisland" functionality work as expected. Name of the test would be "DiffusionWeightedVolumeMaskingTest_removeisland"

For reference, here is the commit message:

//----------------------
BUG: Fix DiffusionWeightedVolumeMasking module and add test. See 0003855

The updated line calculates the magnitude of each gradient, and
determines that an image is "baseline" if the magnitude is smaller than
a threshold.

The use of "abs()" on double values caused the magnitude to
always be 0, and therefore all images were considered as "baseline" and
included in the average computed by the filter and presented as "Output
baseline volume" to the user.

This commit fixes the incorrect logic using "fabs()", simplifies the test
case removing the use of "--removeislands", updates the test harness to use
"compare" functions to validate the output of the module against baseline
images.

Note that the "--removeislands" should be tested in a separate test case.

Co-authored-by: Jean-Christophe Fillion-Robin <jchris.fillionr@kitware.com>

From: Orjan Bergmann <orjan.bergmann@gmail.com>
//----------------------