What are we waiting for to update the photogrammetry database using artificial intelligence?
Yea I find it weird the use of AI hasn’t found it’s way into MSFS yet in any meaningful way with PG and autogen buildings
That said, we’ve all noticed that the photogrammetry quality has been significantly improved in the recent CUs and WUs.
But I agree—AI could probably add real value. If it could just get rid of all those photogrammetry trees in the landscape, that would be huge… or increase the resolution and precision of the ground textures… one can dream
Out of curiosity, where does this image come from? Which AI are we talking about?
Why in the world would you want to add AI into the complexity we are already dealing with at a probable decrease to our performance? It can take some time for AI to analyze an image frame and redraw it
In addition, AI in not free. Someone has to pay for it and it contributes to component price hikes and environmental issues.
No… the way I see it, the AI wouldn’t be used on the PC side in real-time, but rather on the server side for a task designed to enhance the scenery once and for all—not every single time. That way, it wouldn’t place a heavy load on PCs or the scenery streaming process; the streamed scenery would already have been processed long before being sent to the client. Essentially, it would involve upgrading the terrain, photogrammetry, and other objects on the servers. That’s how I picture it…
The test was done with chatgpt; it is certain that using AI in real time is probably impossible, but the current data could be fully processed by an AI to correct the valuable data that has already been collected.
I’m not sure ‘correct’ is the right word when it comes to AI. While I think that it could be used to smooth out the data, some of what it ‘fixes’ may be made up or completely wrong. Whether that matters for this type of data would depend on what is acceptable for each of us, for me, I think it could improve things dramatically and I’m personally not too bothered about 100% accuracy for this type of data.
Mind you, it would take a lot of processing power to ‘correct’ this data and the question then is whether it’s a good use of resources for something that would likely be very intensive for large areas.
MSFS 2020 and 2024 already uses ai to fill in the autogen. And… it does a terrible job.
In my home country, The Netherlands, 80 percent of the autogen houses are just wrong. Filling most Dutch cities with “canal houses” wich are actually only to be found in the center of Amsterdam and some other cities in the western part of our country, is just killing the immersion.
Church buildings all look the same. A generic brown bell tower , and that’s all.
And when flying over the rural part of The Netherlands, the buildings are even worse. It just doesn’t look like The Netherlands at all. It looks more like Scandinavia or Germany.
So, the ai used by MS and Asobo is not doing a great job at all and should be overhauled.
The Autogen used in certain parts is also not of a very high quality either.
I’m pretty certain the current autogen buildings are created algorithmically based on various sources such as OSM. Hence there simply being ‘a church’ where the data says there is one.
Use of the term ‘AI’ these days implies machine learning and I don’t think that’s used in MSFS’s buildings and scenery.
I’ve suggested before that the technology is probably capable, but I don’t think the resources exist to develop and process it.
It will require high polygon geometry with pixel matching algorithms to match at that level which tree leaves, reflections etc at the moment, end up confusing the algorithms and so apparently confuses depth calculations.
So the algorithms scale everything down to low poly geometry- so sayeth our almighty google god anyway! 
This is why you don’t see the name of this company anymore in the opening credits of MSFS2024.
They don’t have a replacement yet, so it seems.
Getting back to photogrammetry, it has improved significantly in the latest City and World Updates. The city now looks completely convincing from an altitude of 500 feet during low-level flight; below that, you start to notice flaws. Eliminating the photogrammetry trees would make it even better. The level of detail is much higher than before.
To balance things out, we really need a quality boost for the autogen as well—that would be great.
Yes, the quality of photogrammetry has increased when flying higher then 500ft in a winged aircraft. But when flying helicopters it isn’t that spectacular anymore.
I’d have to respectfully disagree with you there. I mostly fly Helicopters in only PG areas and I have to say the last WU20 Japan, WU21 Australia and the most recent CU’s the PG quality is so good! Take Adelaide for example or Canberra, both are fantastic in Helicopters, but when you get below 500ft the PG will always look a mess.
PG was never meant to look perfect at 50ft but once you get a bit of altitude, to me at least the PG looks fantastic.
500 and 1,000 feet is a pretty normal en-route low altitude in a helicopter and unless you can do so safely and comply with regulations you’re not going to be flying lower than that.
I just tell the authorities that my religion forbids the use of phones, so I never get “the call” 
I fly autogenerated Mission hub missions from Miltech or HPG H145 with action pack. Mostly HEMS mission where you do not fly higher then 1100ft. Flying over or Landing on a photogrammetry highway or in a city isn’t a really very immersive experience when everything is just a blurry mess.
Not really much can be done about this other than using handcrafted scenery.