that sounds like an interesting effect that you are seeing there (which is absolutely note intended). Would it be possible for you to share the results, so that we can better understand the blockiness of the output?
On a side note: exporting to jpg is often not a good idea, jpg adds lossy compression, which in turn means that you loose data. Tif should be fine.
I have just tested the jpeg export. I think the problem in the case of the simple segmentation really is the jpeg compression that blockifies it. The problem is that the simple segmentation will assign 0 values to pixels of the first label, 1 to pixels of the second label, 2 to the pixels that are classified the third label, and so on. So the jpeg compression was more designed for natural images with higher dynamic range, and produces these kind of artifacts here, where all the information is distributed in only low pixel values, that are mostly organized in larger areas. Can you maybe try exporting to tiff, or png to check the result? Those don’t add lossy compression.
Alternatively you could export to hdf5
edit: added image:
is this the kind of blocking you are seeing?