” By random chance, I satisfied somebody from Microsoft who was looking for a studio to help them out on the brand-new Flight Simulator. The core concept of the new Flight Simulator simulator was to use Bing Maps as a playing field, as a map, as a background,” Putz explained.
Bing Maps photogrammetry data only yielded precise 1:1 replicas of 400 cities– for the vast majority of the world, though, that data doesnt exist. Microsoft and Asobo Studios needed a system for constructing the rest.
This is where Blackshark is available in. For Flight Simulator, the studio reconstructed 1.5 billion buildings from 2D satellite images.
Now, while Putz says he fulfilled the Microsoft team by opportunity, theres a bit more to this. Back then, there was a Bing Maps team in Graz, which developed the first video cameras and 3D versions of Bing Maps. And while Google Maps won the marketplace, Bing Maps actually beat Google with its 3D maps. Microsoft then launched a proving ground in Graz and when that closed, Amazon and others can be found in to get the local talent.
” So it was easy for us to fill positions like a PhD in roof restoration,” Putz said. “I didnt even understand this existed, however this was precisely what we needed– and we found two of them.
” Its easy to see why rebuilding a 3D building from a 2D map would be tough. Even finding out a structures precise outline isnt simple.
Image Credits: Blackshark.ai
” What we do generally in Flight Simulator is we look at locations, 2D areas and then discovering footprints of buildings, which is really a computer vision job,” said Putz. “But if a structure is obstructed by a shadow of a tree, we really require maker knowing due to the fact that then its unclear anymore what is part of the structure and what is not due to the fact that of the overlap of the shadow– however then machine learning finishes the remaining part of the building. Thats an incredibly basic example.”
While Blackshark had the ability to rely on some other data, too, including photos, sensor data and existing map information, it needs to make a decision about the height of the building and a few of its attributes based on very little info.
The apparent next problem is figuring out the height of a building. To identify the height of a building based on a shadow, you require the time of day, though, and the Bing Maps images arent really timestamped. For other usage cases the company is working on, Blackshark has that and that makes things a lot simpler.
Image Credits: Blackshark.ai
” Machine learning takes a slightly different road,” kept in mind Putz. “It likewise takes a look at the shadow, we believe– due to the fact that its a black box, we do not truly know what its doing. Likewise, if you look at a flat rooftop, like a skyscraper versus a shopping mall. Both have mainly flat roofs, but the roof furniture is various on a skyscraper than on a mall. This assists the AI to find out when you label it the best method.”
And then, if the system understands that the average height of a shopping mall in an offered area is typically three floors, it can work with that.
Something Blackshark is very open about is that its system will make mistakes– and if you purchase Flight Simulator, you will see that there are apparent mistakes in how a few of the buildings are put. Putz told me that he thinks one of the hardest difficulties in the project was to convince the businesss development partners and Microsoft to let them use this method.
” Youre talking 1.5 billion structures. At these numbers, you can not do standard Q&An any longer. And the conventional finger-pointing in like a level of Halo or something where you state this pixel is not good, fix it, does not really work if you develop on a statistical basis like you finish with AI. So it may be that 20% of the structures are off– and it actually holds true I think in the Flight Simulator– however theres no other method to tackle this challenge because outsourcing to hand-model 1.5 billion structures is, simply from a logistical level and also budget plan level, not achievable.”
In time, that system will likewise improve, and because Microsoft streams a lot of the information to the game from Azure, users will undoubtedly see changes over time.
Image Credits: Blackshark.ai
Identifying, though, is still something the team has to do just to train the design, and thats actually an area where Blackshark has made a lot of progress, though Putz would not say too much about it because its part of the companys secret sauce and among the primary reasons it can do all of this with practically 50 people.
” Data labels had actually not been a priority for our partners,” he said. “And so we used our own live labeling to essentially label the whole world by 2 or three guys […] It puts a really powerful tool and interface in the hands of the information experts. And essentially, if the information expert wants to identify a ship, he tells the discovering algorithm what the ship is and after that he gets immediate output of spotted ships in a sample image.”
From there, the expert can then train the algorithm to get back at better at finding a particular object like a ship, in this example, or a mall in Flight Simulator. Other geospatial analysis business tend to concentrate on particular niches, Putz likewise kept in mind, while the businesss tools are agnostic to the type of material being examined.
Image Credits: Blackshark.ai
And thats where Blacksharks larger vision can be found in. Due to the fact that while the company is now getting acclaim for its work with Microsoft, Blackshark likewise deals with other companies around reconstructing city scenes for self-governing driving simulations, for instance.
” Our bigger vision is a near-real-time digital twin of our planet, especially the worlds surface, which opens up a trillion use cases where conventional photogrammetry like a Google Earth or what Apple Maps is doing is not assisting because those are simply simplified for images clued on basic geometrical structures. For this we have our cycle where we have been extracting intelligence from aerial data, which might be 2D images, however it likewise could be 3Dpoint counts, which are currently doing another project. And after that we are picturing the semantics.”
Those semantics, which explain the structure in very precise detail, have one major benefit over photogrammetry: Shadow and light info is essentially baked into the images, making it difficult to relight a scene reasonably. Because Blackshark knows everything about that constructing it is building, it can then also put windows and lights in those structures, which produces the surprisingly sensible night scenes in Flight Simulator.
Point clouds, which arent being used in Flight Simulator, are another area Blackshark is concentrating on today. Point clouds are extremely difficult to check out for people, particularly as soon as you get extremely close. Blackshark uses its AI systems to examine point clouds to discover out the number of stories a building has.
” The entire company was established on the idea that we need to have a huge benefit in innovation in order to get there, and specifically originating from video games, where big productions like in Assassins Creed or GTA are now hitting capacity limits by having thousands of individuals working on it, which is extremely difficult to scale, really tough to handle over continents and into a prompt provided product. For us, it was clear that there need to be more automatic or semi-automated steps in order to do that.”
And though Blackshark found its start in the video gaming field– and while it is working on this with Microsoft and Asobo Studios– its really not focused on gaming but instead on things like self-governing driving and geographical analysis. Putz kept in mind that another fine example for this is Unreal Engine, which started as a game engine and is now everywhere.
“And when you look at simulators, from military simulators or commercial simulators, they constantly kind of appearance like shit compared to what we have in driving video games. I think Blackshark is one of those examples for making this possible.”

Microsofts new Flight Simulator is a technological marvel that sets a brand-new requirement for the genre. However to recreate a world that feels alive and real and consists of billions of structures all in the right spots, Microsoft and Asobo Studios relied on the work of numerous partners.
One of those is the little Austrian startup Blackshark.ai from Graz that, with a group of just about 50 individuals, recreated every city and town all over the world with the assistance of AI and massive computing resources in the cloud.
Ahead of the launch of the brand-new Flight Simulator, we sat down with Blackshark co-founder and CEO Michael Putz to discuss dealing with Microsoft and the companys wider vision.
Image Credits: Microsoft
Blackshark is in fact a spin-off of game studio Bongfish, the maker of World of Tanks: Frontline, Motocross Madness and the Stoked snowboarding video game series. As Putz told me, it was really Stoked that set the business on the way to what would end up being Blackshark.
” One of the first games we carried out in 2007 was a snowboarding video game called Stoked and S Stoked Bigger Edition, which was among the very first video games having a complete 360-degree mountain where you could utilize a helicopter to fly around and drop out, land all over and go down,” he explained. “The mountain itself was procedurally built and explained– and also the placement of obstacles of plants, of little animals and other snowboarders had been done procedurally. We went more into the racing, shooting, driving genre, but we still had this concept of positional placement and descriptions in the back of our minds.”
Bongfish returned to this concept when it worked on World of Tanks, merely due to the fact that of how time-consuming it is to construct such a huge map where every rock is positioned by hand.
Based upon this experience, Bongfish started developing an in-house AI team. That group used a variety of machine-learning techniques to construct a system that could find out from how designers develop maps and then, at some point, develop its own AI-created maps. The group in fact ended up using this for a few of its tasks prior to Microsoft entered the picture.

” What we do essentially in Flight Simulator is we look at locations, 2D locations and then finding out footprints of buildings, which is in fact a computer vision task,” stated Putz. “But if a building is obstructed by a shadow of a tree, we really need machine knowing because then its not clear anymore what is part of the structure and what is not due to the fact that of the overlap of the shadow– but then maker learning completes the staying part of the structure. To determine the height of a structure based on a shadow, you require the time of day, however, and the Bing Maps images arent actually timestamped. It might be that 20% of the structures are off– and it actually is the case I think in the Flight Simulator– however theres no other method to tackle this difficulty due to the fact that outsourcing to hand-model 1.5 billion structures is, simply from a logistical level and also spending plan level, not workable.”
Blackshark utilizes its AI systems to analyze point clouds to discover out how lots of stories a building has.


Please enter your comment!
Please enter your name here