In this screenshot we clearly see double shadows, one baked into the Bing image and one generated by the sim. 

I'm curious whether Azure AI can eventually remove baked shadows in Bing images, or at least make them pale enough to be much less noticeable. I'd imagine this kind of double shadow scene can be quite jarring when flying over skyscrapers, mountains and hills etc. 

 

Edited February 19, 20206 yr by FlyIce

the AI is probably confused by how big those shadows are. Its done a pretty good job elsewhere

if Ai cant do it the other option is to manually remove the shadows...

lol j/k you would need thousands of years to finish the whole earth 😉

MSFS has a binary star system modeled!

As far as I know, the AI generates the height of autogen buildings by analyzing the building shadow. So, the AI need to study the baked shadow first and then need to intelligently remove the baked shadow from the map... seems like a lot of work to the AI. 

We have seen a lot of pics.  Has any other pic or video shown double shadow?

Edited February 20, 20206 yr by GlideBy

10 hours ago, Oliver Ooi said:

As far as I know, the AI generates the height of autogen buildings by analyzing the building shadow. So, the AI need to study the baked shadow first and then need to intelligently remove the baked shadow from the map... seems like a lot of work to the AI. 

That is Genius!! And is it really too much work to the AI, nah that's what it's built for right? 😁

I think we still misunderstand what the AI is doing for MSFS but I could be mistaken.

I don't think the AI removes any shadow in the images, just like I don't think it retouches imagery with different color tones... 

The AI simply detects zones of interest (footprints on the imagery like infrastructure, vegetation, color discrepancy zones, shadows etc.).

It collects "smart" data to be processed by Asobo for multiple purposes using most probably procedural tools they've programmed to generate and estimate infrastructure (footprints, types, height, roof types etc), vegetation etc

The data could also feature zones where the imagery colors have radical discrepancies in specific zones. Shadows could be one of these.

The AI learns and uses the learning to work on specific tasks.

In other words, and in my opinion, the AI job is to analyze the pixels variations (in a super fancy way) but just like the algorythms of Adobe Photoshop selection tools do.

Again the AI process is super automated and most probably does it in a way that is more sophisiticated by learning and guessing things like the tolerance between pixels and other factors that in Photoshop a human input is required for evey case.

Now removing the shadows in the imagery, just like balancing imagery zones that have radical color or hues shifts, requires another level of intervention by additional tools with fancy algorythms also similar to what Photoshop features for image adjustments and finetuning.

And this additional color correction intervention, in my opinion, is up to Asobo and MS to decide if they do it or not.

But I don't think this task is part of the AI in my opinion.

 

Edited February 20, 20206 yr by Claviateur

Some things will either be handled by AI or not handled at all. It's impossible to manually remove shadows across the entire globe, but so far, MSFS has done a decent enough job at it. 

If you see some double shadows, it is what it is. This isn't going to be a perfect experience, just a far improved one. 

Edited February 20, 20206 yr by bonchie

6 minutes ago, bonchie said:

Some things will either be handled by AI or not handled at all. It's impossible to manually remove shadows across the entire globe, but so far, MSFS has done a decent enough job at it. 

If you see some double shadows, it is what it is. This isn't going to be a perfect experience, just a far improved one. 

Well, right now photogrammetry isn’t all the globe. Are 400 cities. And no so many tall buildings with double shadow like this. 
 

It could be improved, or not, we will see. But saying it is what it is, is asume too much on alpha stage. 

15 minutes ago, aleex said:

Well, right now photogrammetry isn’t all the globe. Are 400 cities. And no so many tall buildings with double shadow like this. 
 

It could be improved, or not, we will see. But saying it is what it is, is asume too much on alpha stage. 

And I think based on my observations that the Photogrammetry areas were not processed like the rest of the (procedural) world. It seems as if the Photogrammetry area/data was implemented almost as is. It looks beautiful thanks to the rendering capability of the engine...

Azure AI is in a constant learning process I wouldn't worry at all about it, but good to be mentioned during the testing phase...

Look for Giza pyramids screenshoot, they have a nice shadows, hard to tell if it's generated by the sun.

In my own opinion they are generated by the sim and there is no signal of originals shadows.

5 hours ago, Claviateur said:

But I don't think this task is part of the AI in my opinion.

We know that when Bing maps are updated, that the sim terrain, buildings, etc., will also be updated.  We know that Asobo doesn't want to do things by hand in the future.  So we can assume that these Bing map changes will be handled automatically.

Next, we know that for certain areas it knows the shadow dimensions, as it uses it to calculate the building height.  So if Azure, the AI platform and more, just passes that on.  Then that means the client would have to erase shadows.  And every client in the world would have to do it over and over.  This seems like a waste to me.  Why not just do it once and be done with it?

If your point is that maybe it is not AI, but instead an algorithm that runs on Azure, then I think we are mincing words.  As AI is sometimes just an algorithm, think of enemy AI in most games. 

1 hour ago, GlideBy said:

We know that when Bing maps are updated, that the sim terrain, buildings, etc., will also be updated.  We know that Asobo doesn't want to do things by hand in the future.  So we can assume that these Bing map changes will be handled automatically.

Next, we know that for certain areas it knows the shadow dimensions, as it uses it to calculate the building height.  So if Azure, the AI platform and more, just passes that on.  Then that means the client would have to erase shadows.  And every client in the world would have to do it over and over.  This seems like a waste to me.  Why not just do it once and be done with it?

If your point is that maybe it is not AI, but instead an algorithm that runs on Azure, then I think we are mincing words.  As AI is sometimes just an algorithm, think of enemy AI in most games. 

Sorry english is not my native language. Pardon 🙂 

I already metioned in one post on this forum that for my scenery work for XP11, I use a similar process (on paper) to create VFR autogen procedurally.

I use OSM data and I use a 3rd party application to process it and create Autogen (infrastructure, roads, powerlines, forests, parks, palces of worship, malls, factories, forests etc etc ). I write rules in an XML files that the 3rd party tool reads and builds the autogen for that area accordingly. 

The OSM data is similar to the data the Azure AI creates by analyzing the Bing imagery, but OSM is manually done by people.

Whereas the algorythms I was reffereing to, are part of the 3rd party tool I use to process the data with its XML rules.

This process creates 3D autogen.

A similar process using another tool (or maybe the same super tool does it all at Asobo, I do not know) but that other tool can process imagery.

Maybe I am mistaken, the AI does not only fetch the data (similar to OSM) but has image processing and 3D generating capability in its deep learning process. 

So all in all, my vision of the process is that the Azure AI fetches the data but does not create autogen or retouch imagery.

Again, I could be mistaken.

Edited February 20, 20206 yr by Claviateur