To learn where in the world Guyana or Myanmar are, all you need to work with is maps that are aligned so that North is toward the top of the screen.
Just like to learn algebra, you don't need to solve equations upside down, or to learn to read, you don't need to read upside down. These variations don't speak to the semantics of the main activity whatsoever.
If you know where every country is from doing spaced repetition and someone gives you a world map upside down, you will easily find a given country; you can mentally rotate the map then.
I think parent is more concerned with the way that after enough times seeing the same image and linking it to an answer, your brain does what your brain does best, which is find mental shortcuts. Gradually you come to associate the image with the answer, without regard to any kind of "why". This process happens rather quickly, and when you're faced with even only a slightly different image, you suddenly find it's very difficult to remember what the country is, despite blazing through it in your reviews.
Changing the image slightly helps force you to actually think about _why_ you know the image is depicting the country in question, rather than simply building automatic visual recall of a country name tied to a specific image.
You can of course achieve the same thing with just having a bunch of different "stylesheets" for your map or something, but that probably requires manual creation in order to not be a complete mess, and still probably isn't as good for building transferable knowledge. This approach on the other hand is much better at building transferable knowledge, but requires more frustration and a much longer time horizon to be able to usefully utilise the knowledge.
If you're, for example, using spaced repetition to learn a complicated foreign script, which can be written in numerous ways (numerous printed fonts, and handwriting styles), you have to be able to recognize and name these variants. That is well and good.
If we are training name to location for a set of countries, the front of the card is the name. You only need multiple versions of that if the country name has spelling or naming variations.
If you're going from map shape to naming the country, and you want to be able to do that in any orientation, then you benefit from training on map rotations. I'm guessing four rotations ought to be enough: upside-down and ninety degrees CW and CCW. That already quadruples the training set; there has to be some reasonable balance.
Just like to learn algebra, you don't need to solve equations upside down, or to learn to read, you don't need to read upside down. These variations don't speak to the semantics of the main activity whatsoever.
If you know where every country is from doing spaced repetition and someone gives you a world map upside down, you will easily find a given country; you can mentally rotate the map then.