I thought I would show the sort of issues I come across using the SDK as far as tree placement is concerned and why I complain about the way it currently works.
2 shots: The first shows how the sim misinterprets the ground and puts trees galore all over the ortho. It does this on just every mountain in Cumbria and its reasonable to assume it does this elsewhere too.
The second shot shows what it should look like after an exclude has been added to hide the trees that are being placed incorrectly.
So to me this is so inaccurate that the developers should take another look at it. We are not talking about a few leaves being wrong on a tree here, its tens of thousands of trees per square mile that should not be there and completely changes the landscape so that its no longer recognisable if you know the area.
Some of you may think that this is OK and thatâs fine, that is your prerogative but for those of us who are looking for a reasonably accurate portrayal of the world, this sort of issue is a problem and brining it to peoples attention so that maybe we can get it corrected in no bad thing in my opinion.
Hard to see from these shots as I canât expand them. I donât know if this is something you can tell from the image whether trees should be there or not or whether you need to have local knowledge to know that.
If you look at the top shot Gordon you can see a forest of light green trees on the left hand side of the image that are not present in the bottom shot. So what you are seeing is a load of trees that the sim is placing because it is not identifying the difference between trees and ground foliage.
I have raised the issue via zendesk but I wanted people who think I am exaggerating to see a vivid example of what I am coming across.
All the larger dark blobs are trees and shadows by the way. The image below is clickable and is what the area actually looks like.
The problem with the slider, (I dont know if that has an effect on misplacede trees or not) is that even if it corrects the issue its almost certainly going to have an effect on tree density elsewhere too. Ideally what is needed is for the AI to learn how to recognise the difference between trees and foliage. Bracken seems to be a particular problem for the AI.
It is in other areas in AI, not just trees. Near my home are several oil tank âfarmsâ. These large tanks, left behind after two refineries were demolished, are displayed as round houses, complete with doors, windows, and round peaked roofs. The whole AI program, although a novel and good idea, will probably take some time to mature.
Thanks, the bigger image makes it much easier to see what you mean. I suspect this is a balancing act on the basis that theyâve got to apply a procedure here that is able to work across the world (or at least in large regions, as I think from the famous tree height mod, that became adjustable by region and biom). It could be that if they tune this so that trees donât appear here, then there are way too few trees elsewhere. No idea how granular they can get on this, but if you donât know the areaâŠ
It would seem to me that there are always going to be inaccuracies given the global coverage. What would be helpful would be a simple mechanism to be able to raise scenery bugs/inaccuracies with Asobo - other than zendesk - or (has been suggested elsewhere) some way of the community editing the data and going through some validation process with Asobo to be adopted into the full sim for everyone over time.
I think people are placing too much expectation on AI. Asobo is covering the entire world with what most would argue is a pretty reasonable and for the most part an extremely realistic portayal of the planet. While Iâm sure they can continue to work on the hueristics and tune the results and probably augment with other data sources like satellite foliage maps, its unlikely to think it will ever be perfect. In fact, its more likely that they will reach a point where fixing something in one section of the world has adverse effects in another (as evidenced by the tall buildings caused by the Japan patch).
An alternative way to reach a satisfying conclusion is to adopt a gateway scenery system. There are far more of us who have both vested interest and local knowledge of what things âshouldâ be. Continue to make the scenery editor easier to use, find an organized way to gather and vette results and incorporate the good (and legal) stuff into future global updates.
I look at the AI pathways as a great way to âseedâ the system. We are, afterall, starting from a pretty great place out of the box. But there are a lot more users than there are staff at Asobo - why not unleash the masses and capitalize on the results - for all of our benefits.
