I tried to run the application on big images, and also ran into error.
After running each step of the LSMS workflow workflow, it looks like step 3 and 4 both have a huge memory consumption when dealing with a large number of segments. I need to look at the code in depth to find out what’s happening.
As an alternative you could try to use the Segmentation application instead, using the meanshift algorithm, and the vector mode (the raster mode does not handle big image). Note that the results will differ from what
LargeScaleMeanShift would produce. In this case the input image is divided and segmented in tiles, instead of doing the segmentation on the whole image. And then there is a
stitching optional step that tries to merge polygons that might have been split due to the tiling process based on distance between the mean radiometry of the segments.
Hope that helps