How good is Azure?

[jlund 380]

We have all seen videos where there's a solid "wall" under bridges. Yes it's still beta, BUT how good is Azure ai? Will it be possible for it to learn that it's a bridge and it's not supposed to be solid, or does Asobo/MS have to correct it manually? It will be nearly impossible to check all bridges in the world, so can it eventually be fixed by azure? And the same goes with ie. container cranes, they also seems pretty solid, so it may take some time before it's fixed, just wondered if it's even possible for a computer programme to figure out what's what.

BTW. I haven't yet seen any videos with wind turbines (windmills), if you have, can you please post a link to a video with it?

[highflyer2020 607]

Well it's AI...so I assume once you train it to see bridges better it will get better at it over time but don't confuse "better" with perfect. 

[MattNischan 3,778]

It's an extremely difficult programming challenge, even with a good machine learning algorithm.

Bridges come in all different types, the difference between which are basically nearly impossible to discern from imagery data. I could see training the algorithm to learn that bridges are road looking things over water looking areas, but then what makes a bridge over a land looking area different from a road over a land looking area, if you're staring at it straight down? Raw road map data doesn't always even discern these things terribly well either, if you're using that as a backstop to your algorithm.

I imagine over time they'll improve and add some kind of stock non-solid supports to various bridge types, but it's a tricky problem to solve. The supports have to follow the underlying terrain and have to be plausible enough, but also generic enough that you aren't building full 3d model data for each bridge in the world.

As an aside, Azure is simply the cloud architecture that the machine learning based Bing data ingestion is running on. They're calling it Azure AI for short, but it's just machine learning algorithms running on machines hosted in the Azure cloud. There's no special AI platform service provided by Azure. So, it isn't about what Azure, the cloud hosted computing platform can do so much as it is about how well they can refine the custom software that's running on top.

[Jeff Nielsen 1,386]

I think we might learn more about this in the BlackShark AI partnership series?

[Slides 1,552]

Azure is just the cloud platform. Blackshark is the AI for generating the 3D terrain. 

[EzRyder 35]

i have seen several youtube vids where that seems to have been fixed, or at least on the bridges im seeing now
there is still some roads going 'under/thru' rivers but i think they may have already fixed the solid bridge thing

edit: watch any of the selected content creators, not the earlier alpha vids

Edited August 4, 2020 by AirWayMan

[omarsmak30 4,068]

Plus with others said in regards to Azure and BlackShark AI, the quality of the trained model makes a lot of difference. A machine learning model is a model that has been trained for repeated pattern, continues learning to update the data points in the model. Hence, at first stage, the model won't be perfect, which explains sometimes the generated autogen may be odd. However, as I mentioned is a continuous learning effort to improve the model, which as result, I think we will see in the future improved autogen based on this feedback.