This is something I can’t wrap my head around. In the partnership video for blackshark.ai they explain how they extract the building footprints from top down imagery and then use various parameters to “paint” the facades based on how buildings in a certain area usually look. Even though they don’t mention it I was under the impression that they also used data from Bing’s Bird’s Eye View where available. I base that on the fact that some cities look almost photo realistic and that seems impossible if they only used top down imagery.
There you can drag the map to any location and look at the Bird’s Eye View imagery where available. I was very surprised to see that some cities that have that photo realistic quality in the sim aren’t covered! Take Copenhagen as an example, looks stunning in the sim but Bing only have a few small areas in the outskirts of the city covered by Bird’s Eye View.
This makes me wonder if they are actually making these locations manually? If not, what data are they feeding their AI with in order to generate that photo realism in some areas?
The reason Copenhagen looks that good is that it has photogrammetry, which has nothing to do with bird‘s eye view. I‘m not sure if bird‘s eye view is even used for MSFS.
There are two ways of making scenery in the sim. One is what you have mentioned, AI-driven autogen scenery. The second method is through photogrammetry.
Bird Eye is not photogrammetry. Satellite images or even aerial photos are not enough to make fully 3d scenery unless they are taken from many different angles from planes, helicopters or drones.
There is a 3rd way that they have not explored yet. It is technically possible to extract 3d buildings (just buildings) from very high-quality satellite images, but it seems that it has not been done in MSFS and I hope it will be done some time down the road. Bird’s Eye quality seems to be good enough for that technique.
Thank you for taking your time to explain this, first link was very nice!
However, I can’t stop thinking that it should be possible to somehow use the imagery from Bird’s Eye View to improve the accuracy of the AI generated buildings. These are very sharp pictures taken from four different directions. Without knowing much about AI and the algorithms involved I’m guessing it wouldn’t be impossible to use these to generate facades that where closer to the real ones. Not as close as with photogrammetry, but closer than by just using cultural and whatever parameters they use. But that’s a whole other story.
The second video I posted is exactly that. Buildings generated from high resolution images. That is not being used in MSFS yet, so there is my hope we will see it leveraged some time.
