A nice summary of the infrastructure types in MSFS

[Claviateur 789]

This screenshot seems to summarize nicely the 3 types of infrastructure we could have in MSFS

• Extruded footprints, similar to what we call "Facades" in the other simulator. 
• Generic (ready made) 3D library infrastructure (classified per region) placed procedurally. They are those with more details on the roofs I suppose , and don't match the exact footprint.
• Custom 3D landmarks

Edit: Ok now I see what looks like extruded footprints with details on the roofs ummm

 

Edited May 1, 2020 by Claviateur

[Claviateur 789]

Thanks to Shack95, who identified the screenshot location, I noticed the following:

The infrastructure rendered in MSFS, is also based on pseudo-footprints that cover multiple infrastructure surfaces in real life.

Here for example, the buildings in MSFS cover multiple smaller infrastructure in real life.

It could be that the procedural algorithms during the implementation of the infrastructure make this footprint grouping happen.

Or the scanning of the aerial images produce this type of grouping / simplification... Maybe but I doubt...

Edit: Or these are ready made 3D library objects selected with a procedural rule to match as much as possible the empty zone to cover.

But we see that it's not usually one for one when it comes to footprint matching.

Also as for now, the height of the infrastructure is procedural when metadata is not available in the footprints...

Edited May 1, 2020 by Claviateur

[Darcanlos 89]

It still is very believable for non photogrammetry, unless you live there, of course. I think it does a pretty good job taking into account it's all realtime and included from the get-go. 

[Claviateur 789]

23 minutes ago, Darcanlos said:

It still is very believable for non photogrammetry, unless you live there, of course. I think it does a pretty good job taking into account it's all realtime and included from the get-go. 

Absolutely! I agree...  if, as a user myself I fly over there, I would not know or would not mind at all. The MSFS result is very interesting.

Yet these analyzes are just part of my reflexes for scenery developement techniques and specs, not to evaluate how realistic it is compared to real life.

[Dominique_K 10,133]

See my comment in Shack's thread. If I am not mistaken, the scenery is seriously wrong

[Claviateur 789]

29 minutes ago, Dominique_K said:

See my comment in Shack's thread. If I am not mistaken, the scenery is seriously wrong

I saw the comment there and I wanted to reply until I noticed your comment here 🙂

As I said, in this thread I am not analyzing or evaluating the realism of the infrastructure vs real life but rather curious about understanding the different global scenery techniques Asobo are using.

Yet this shows that sometimes metadata is not available in the footprints (Ex: type, height etc), and in other situations the footprints are most probably not (yet) available for the whole planet, hence alternative procedural techniques are applied to fill empty zones with 3D infrastructure.

Procedural methods produce plausible results but never exact matches as you know. 

Edited May 1, 2020 by Claviateur

[Dominique_K 10,133]

8 minutes ago, Claviateur said:

I saw the comment there and I wanted to reply until I noticed your comment here 🙂

As I said, in this thread I am not analyzing or evaluating the realism of the infrastructure vs real life but rather curious about understanding the different global scenery techniques Asobo are using.

Yet this shows that sometimes metadata is not available in the footprints (Ex: type, height etc), and in other situations the footprints are most probably not (yet) available for the whole planet, hence alternative procedural techniques are applied to fill empty zones with 3D infrastructure.

Procedural methods produce plausible results but never exact matches as you know. 

My comment was not directed against your own post 😉. More of a complement. 

Asobo's tech fails here as it is not "plausible". Until Schack pinpointed the exact location I was pretty sure that it was Bangkok but I couldn't make up my mind because of these buildings so unBangkok-like .

Clearly, the AI needs to be educated.  

 

[RALF9636 1,902]

59 minutes ago, Dominique_K said:

See my comment in Shack's thread. If I am not mistaken, the scenery is seriously wrong

 

Just to put things into perspective, this is a screenshot posted lately on the avsim screenshot forum, showing the latest P3D version:

 

If this:

is called "seriously wrong" now, I'd be quite happy with a seriously wrong scenery. 😉

Most important for me would be a plausible scenery. And this image looks plausible to me, since I am not familiar with the location.

Of course a scenery which is only plausible but not completely correct can be disappointing for those familiar with the location. But personally I wouldn't mind too much about that. Improving a plausible scenery into a correct scenery can be a domain for third party regional addons or subsequent manual or AI improvements by Asobo.

 

Edited May 1, 2020 by RALF9636

[Claviateur 789]

18 minutes ago, Dominique_K said:

My comment was not directed against your own post 😉. More of a complement. 

Asobo's tech fails here as it is not "plausible". Until Schack pinpointed the exact location I was pretty sure that it was Bangkok but I couldn't make up my mind because of these buildings so unBangkok-like .

Clearly, the AI needs to be educated.  

 

Oh no I did not think you directed your comment against my post at all 🙂

Again, I think we misunderstood what the AI role is here. The AI scans imagery if it is asked to and brings footprints or vegetation zones. These are the roles we are almost certain the Azure AI have because it was done by MS before MSFS for the US and Canada.

Here is the data resulting from the MS AI Scanning process for Canada with explanations:

https://blogs.bing.com/maps/2019-03/microsoft-releases-12-million-canadian-building-footprints-as-open-data

https://github.com/microsoft/CanadianBuildingFootprints/blob/master/README.md

Now let's say the AI was put to scan and fetches footprint data for the rest of the planet (It does not seem like it), I don't think it brought additional metadata or tags with the footprints like Height, Building types etc.

So the AI brings data or does not bring data because it was not put on the task.

The data is used by Asobo to generate global infrastructure with algorithms and rules. In the absence of additional tags, the height, type of infrastructure etc is guessed procedurally.  Now in the absence of footprint data, the whole infrastructure shape and other aspects is guessed procedurally...

Now let's say all data is here + tags but there are no visual assets to make this variation happen (i.e: limited type of architecture for residential blocks / facades textures), then the residential blocks will look similar in Europe and Asia let's say.

The final look and feel of the infrastructure is the result of this whole chain reaction of data + processes and assets availability.

Edited May 1, 2020 by Claviateur

[Dominique_K 10,133]

2 minutes ago, RALF9636 said:

 

Just to put things into perspective, this is a screenshot posted lately on the avsim screenshot forum, showing the latest P3D version:

 

If this:

is called "seriously wrong" now, I'd be quite happy with a seriously wrong scenery. 😉

Most important for me would be a plausible scenery. And this image looks plausible to me, since I am not familiar with the location.

Of course a scenery which is only plausible but not completely correct can be disappointing for those familiar with the location. But personally I wouldn't mind too much about that. Improving a plausible scenery into a correct scenery can be a domain for third party regional addons or subsequent manual or AI improvements by Asobo.

 

Ralf,

I do not compare with p3d but from what I know of Bangkok which is confirmed by  what Google Earth Street View shows. 

I understand that plausibility is in the eye of the beholde but Bangkok is a major tourist attraction. The Grand Palace with the Emerald Buddha Temple close to it is a large compound downtown  and the city has around  small one or two storey buildings along busy streets which are so typical. 

This is not to be negative  to report a thing may help Asobo to release a better FS20.  On the contrary. 

[Claviateur 789]

2 minutes ago, Dominique_K said:

Ralf,

I do not compare with p3d but from what I know of Bangkok which is confirmed by  what Google Earth Street View shows. 

I understand that plausibility is in the eye of the beholde but Bangkok is a major tourist attraction. The Grand Palace with the Emerald Buddha Temple close to it is a large compound downtown  and the city has around  small one or two storey buildings along busy streets which are so typical. 

This is not to be negative  to report a thing may help Asobo to release a better FS20.  On the contrary. 

I see what you mean but in this case the city must be created as a custom scenery as we can't count on such a global implementation process to ensure such accuracy unfonrtuately.

 

[Dominique_K 10,133]

I don't disagree but we see the limits of the AI for the first time. It obviously needs to be "educated" . Get the temple out and you have city which could be in Eastern Europe. No good. 

Some months ago, the team said they had recruited a Ph. D. in roof reconstruction, I don't know whether that was  a joke or not, mais they definitely need to get a  specialist of Asia to train the AI, maybe an architect in Bangkok, Taibei or Singapour .

 

 

[RALF9636 1,902]

2 hours ago, Dominique_K said:

Ralf,

I do not compare with p3d but from what I know of Bangkok which is confirmed by  what Google Earth Street View shows. 

Just to clarify the obvious: The P3D shot isn't Bangkok but San Francisco of course. I just posted it to display the general scenery quality we have so far. And I think to have low quality textured skyscrapers standing in the middle of roads really is "seriously wrong"...

 

Quote

I understand that plausibility is in the eye of the beholde but Bangkok is a major tourist attraction. The Grand Palace with the Emerald Buddha Temple close to it is a large compound downtown  and the city has around  small one or two storey buildings along busy streets which are so typical. 

This is not to be negative  to report a thing may help Asobo to release a better FS20.  On the contrary. 

Of course accuracy would be better than plausibility as the end result (and the more famous a location is the more important that would be). But I don't think we can expect global accuracy from the AI in the foreseeable future. 

So these famous places will need to be custom made (if not covered by photogrammetry) which will take a lot of time and effort for a global scope. I would be fine if that would come subsequently be it by Asobo or third parties.

Of course there's nothing wrong with reporting shortcomings to improve the sim and I didn't meant to say that. But getting too perfectionist will delay the release - and I really can't wait for MSFS to be released. 😉

 

Edited May 1, 2020 by RALF9636

[Claviateur 789]

17 minutes ago, Dominique_K said:

I don't disagree but we see the limits of the AI for the first time. It obviously needs to be "educated" . Get the temple out and you have city which could be in Eastern Europe. No good. 

Some months ago, the team said they had recruited a Ph. D. in roof reconstruction, I don't know whether that was  a joke or not, mais they definitely need to get a  specialist of Asia to train the AI, maybe an architect in Bangkok, Taibei or Singapour .

 

 

Well, again, I am not sure the AI fetched the basic shapes of the planet footprints, let alone other data like type of building, height, area type etc. This is in my opinion too complex. 

And then the visual assets (i.e: facades textures + roofs) must be created for every world architecture type.

The Ph D. role might have been put on ice considering the situation or might be working as we speak on what he/she needs to do.

As long as we are in the Alpha process, things could be added or refined as we saw in the tree types, it was not an AI thing but rather the application of the rules to produce the results or the availability of the visual assets.

Hehe I liked the "mais" in the english sentence 🙂

 

Edited May 1, 2020 by Claviateur

[Gulfstream 416]

Microsoft says they are sharing their building footprint data with OpenStreetmap.

Has there been any confirmation that user-submitted buildings on OSM may be used in the simulator?  This would allow for proper building heights where the algorithm otherwise wouldn't be able to tell from a top-down view.

I'm thinking of areas that do not have aerial 3D photography, it's hard to know the heights of buildings gathered from an algorithm that only sees rooftops.