A nice summary of the infrastructure types in MSFS

[Claviateur 789]

8 minutes ago, Kopteeni said:

I have the same feeling but I don't think it's going to be the Avsim crowd. On Reddit and youtube there seems to be a number of people who are basically expecting a GTA 5 with planes at this point 😬

I think the expectations come down to the misunderstanding of what's the role of Azure AI in this project

I believe that many thought the MS AI creates scenery based on deep or machine learning whereas its role is scanning imagery and fetching data if requested...

[versus 61]

I hoped that MS could do even better than Orto4XP + X-Europe OSM (those familiar with X-PLANE 11 will know what I am talking about) using some advanced AI algorithm applied to Bing satellite imagery that could be a better presentation of 3D objects than OSM data and its application through OpenSceneryX libraries. Shapes and height being one part of the picture and  facades and  textures the other. 

Would be really interesting to have a non-photogrammetric screenshot  from FS2020 with lots of objects compared to Ortho4XP  + X-Europe and, of course, RL photo, to see if  AI is only a PR stunt or it goes a step further than OSM. In this regard underlying satellite imagery is only important as a possible starting point for processing and not for product comparison. Of course, if it is strictly procedural, it can prove more accurate in areas  with scarce OSM coverage.. 

 

[Claviateur 789]

20 minutes ago, versus said:

I hoped that MS could do even better than Orto4XP + X-Europe OSM (those familiar with X-PLANE 11 will know what I am talking about) using some advanced AI algorithm applied to Bing satellite imagery that could be a better presentation of 3D objects than OSM data and its application through OpenSceneryX libraries. Shapes and height being one part of the picture and  facades and  textures the other. 

Would be really interesting to have a non-photogrammetric screenshot  from FS2020 with lots of objects compared to Ortho4XP  + X-Europe and, of course, RL photo, to see if  AI is only a PR stunt or it goes a step further than OSM. In this regard underlying satellite imagery is only important as a possible starting point for processing and not for product comparison. Of course, if it is strictly procedural, it can prove more accurate in areas  with scarce OSM coverage.. 

 

The theoretical process of MSFS global scenery is just like the process you mentioned: Orhto4XP scenery + World2XP (OSM driven) scenery (that creates X-Europe type of global scenery).

This is on paper because the refinement of the aerial imagery (i.e: decals to remove blurs or other methods) and the algorithms with rules and filters could be more sophisticated at Asobo 

The AI is not a PR stunt, the AI brings additional data to couple with Bing + OSM + XY data. As simple as that.

All these data sources are processed to create scenery that is in the same style of X-Euopre or World2XP procedural scenery.

But then the visual assets of MSFS including volumetric trees + the power of the engine make the whole thing renders way better than the global scenery you mentioned for the other sim..

 

Edited May 5, 2020 by Claviateur

[RALF9636 1,955]

1 hour ago, Claviateur said:

I think the expectations come down to the misunderstanding of what's the role of Azure AI in this project

 

I agree. I have no idea how Azure AI works in detail and I only understood a fraction of what you explained in this thread 😉

But how I understood Azure AI from the beginning is that it is meant to get data from different sources (images, "footprints", OSM, whatever...) and use these data to create a plausible representation of the scenery worldwide. The operative word here is "plausible", which is not "accurate". 

Some seem to expect a worldwide accuracy by default created by the AI. I don't think this is going to happen in the foreseeable future. Total accuracy will always take some manual editing. That cannot be done on a global scope. Still I have no doubt there will be lots of third party developers who will create more accurate regional addons.

And I am perfectly fine with that. I think it is fantastic what the Azure AI can already do to create a worldwide scenery which looks most convincing all around the globe. There might be a building where it should be a wall, buildings too small or too large, a banyan tree when it should be a palm or even a wrong colour of a roof. So what? I want to fly above the scenery mostly at minimum 1000 ft with at least 100 kts and won't take notice of every single object on the ground. Most important for me is the overall plausibility of the scenery, creating the immersion of being there when flying an aircraft at an adequate altitude and speed. I am not interested in shooting still images of an accurate virtual world to compare them with reality. 

That doesn't mean that the AI cannot or should not get better over time. But let's not forget where we are coming from. This is still supposed to be a flight simulator, not a virtual earth. And we all know how the flight simulators we have now look compared to what we have seen from MSFS so far.

 

Edited May 5, 2020 by RALF9636

[MatthewS 761]

7 hours ago, Claviateur said:

And below, it seems the long infrastructures are part of a Palace tag. But this tag does not do much for such a scenery generation process hence, the rules and filters in the scenery generation script produced long residential buildings for this area... Yet this is something to expect for a lot of areas on this planet where footprints are approximate... 

Unfortunately Asobo won't be able to do much with such a global scenery based on rough Data. 

Have you ever seen such long and thin buildings in real life? I've not.   If they do exist they would be very rare IMO.

Asobo should filter data like this so that OSM "buildings" with long and thin footprints are not rendered in game as monolithic apartment blocks, instead just render random 1 or 2 storey buildings (or nothing) over the footprint.

With that filter alone the skyline of Bangkok would be improved massively.

[Claviateur 789]

1 hour ago, MatthewS said:

Have you ever seen such long and thin buildings in real life? I've not.   If they do exist they would be very rare IMO.

Asobo should filter data like this so that OSM "buildings" with long and thin footprints are not rendered in game as monolithic apartment blocks, instead just render random 1 or 2 storey buildings (or nothing) over the footprint.

With that filter alone the skyline of Bangkok would be improved massively.

Yes I agree about this filter possibility, but then let's see how this will evolve. I see it like the trees people were afraid they were 1 type. More types are appearing in every Alpha.

If an enthusiast like me can write scripts to create such scenery, highly skilled game developers are able to write all sort of conditions and filters. They have control over their engine and its core capability.

So I am not worried about their skills and ability to do any of this, technically they are highly skilled.... But then MS decides in my opinion if such refinement is a priority.

 

Edited May 5, 2020 by Claviateur

[Tuskin38 6,489]

So I wonder what they do about areas without OSM data, or whatever they use. FSX Style autogen based on the data type?

[Claviateur 789]

2 hours ago, RALF9636 said:

I agree. I have no idea how Azure AI works in detail and I only understood a fraction of what you explained in this thread 😉

Azure AI is like a photo scanner but here it scans  Bing aerial material 🙂 

The rest of my post is for those who are familiar with the other simulator's global world scenery generation using a similar process with 3rd party tools. 

[Claviateur 789]

Just now, Tuskin38 said:

So I wonder what they do about areas without OSM data, or whatever they use. FSX Style autogen based on the data type?

As I mentioned in previous posts, they use procedural techniques. Example: for a specific area, put randomly houses and buildings from this list of objects on one side or another of the roads following a specific pattern, with a specific density and at random angles or alligned...

Something of this sort.

[ahsmatt7 1,199]

On 5/2/2020 at 11:30 AM, Kpeters said:

Omg that looks terrible I don't recall my FSX looking that bad.

I bet it looks the exact same....after all, p3d came from fsx.

[Dominique_K 10,138]

A couple of remarks.

There seems to be a fear that  any criticism could delay the release Not so. The date and conditions have obviously been set in the contract between MS and Asobo. Nothing we can say will change anything. On the other hand, as a long term project, any remark will help. And, as much as I look forward to FS20, it is not a holy cow just because it is an order of magnitude lovelier than my favorite sim of today. 

Plausible vs accurate. Indeed, plausibility is the goal and not accuracy. I do not mind not to see all the temples and palaces in Bangkok or the planetarium in Tokyo.  What I say is to render Bangkok with massive blocks of habitation is not plausible. It looks like the ugly suburbs we have now in France not Bangkok. Not plausible. 

Azure can define the land use in a sat image, pixel per pixel. To which extent Asobo has access to all its functionalities or whether special API are constructed is unknown to me. Assuming that they replicate the methodology used today by the XP folks to make sceneries might be true but it is an assumption not a fact, if I am not mistaken.

 

 

 

[Shack95 2,658]

I checked some more “exotic” and remote places, where there are no OSM footprints. It turns out that all the buildings are there and placed accurately.

Cotahuasi Canyon, Peru:

Near Islamabad, Pakistan:

Dongxingpucun, China:

The same can be seen in the non-OSM areas posted further up by ca_metal and me. What I take from this is that we might indeed have almost every house in the sim and most of them accurately placed and with the correct roof colour. Hight, plausible textures and shape is another question.

One more interesting detail. In last week’s Hong Kong screenshot we can see the Shatin Sewage Treatment Works. Although most of the tanks (encircled) are covered in the satellite image, you can clearly see water texture showing through in the sim. This might suggest that they don’t use image scanning to determine water bodies but rather other sources such as OSM, where these tanks are marked as water. Also the racetrack stables on the left look a bit “OSMish”.

By the way this is not meant as criticism or nitpicking. I’m amazed by what they can do and how it all looks. It’s pure curiosity (and procrastination from work😅).

Edited May 6, 2020 by Shack95

[Dominique_K 10,138]

1 hour ago, Shack95 said:

 

 This might suggest that they don’t use image scanning to determine water bodies but rather other sources such as OSM, where these tanks are marked as water.

Also the racetrack stables on the left look a bit “OSMish”.

 

I went to the ESRI site and Shatin STW covered basins are rendered as water in non satellite imagery. That can explain why the water colors have generally not looked so realistic until now if they don't get them out of the sat imagery.

The racecourse seems to have a mesh problem maybe caused by the slanted spectator stands ?

Fascinating in any case, isn't it ? In all that, one thing is sure, the forums will be inundated by remarks and comments on the accuracy of the landscapes, negative and positive, minutes after release. Maybe Avsim will have to set up a dedicated sub for it 🙂 ! 

 

[Shack95 2,658]

29 minutes ago, Dominique_K said:

Fascinating in any case, isn't it ?

Absolutely. And I‘m still amazed how good  scenery that doesn’t have photogrammetry and isn‘t manually polished up can look. I mean this Shatin screenshot looks absolutely fantastic if you ask me. 

31 minutes ago, Dominique_K said:

the forums will be inundated by remarks and comments on the accuracy of the landscapes, negative and positive, minutes after release. Maybe Avsim will have to set up a dedicated sub for it 🙂 ! 

😂  most probably yes.

[Kopteeni 819]

I think I just found my own spot in the world that I'll be using as a benchmark for their scenery generation. This is Rovaniemi airport (EFRO) with its aerial imagery of 5 different sources (also splitting the airport itself) and lots of clouds:

 

 

Hats off to Asobo if this remote airport and its vicinity in Northern Finland is going to look anywhere close to decent 😄

Edited May 6, 2020 by Kopteeni