ASP writes the integer correlation (D_sub.tif and D.tif) as an image with three bands: x direction disparity, y direction disparity, and mask bit. Locations at which the disparity is invalid have their mask band set to zero. The x and y values at those points though may end up being set to strange values. This is not happening for the refined disparity, RD.tif, where the x and y values are explicitly set. It would be nice if this is cleaned up. Note that this is not an actual bug as it won't affect the final results.