Combining the strengths of our P3D community

[PavlinS 179]

2 hours ago, Rogen said:

Yes of course a scenery fully decked out with objects is the ideal, but as you say that'll take years.

I know. Still, some of us has the patience. Others do not have it and sell out stuff that do not improve almost anything. 

[dave2013 1,140]

You guys have brought up some interesting ideas.

I like the idea of getting rid of the 2D texture/landclass tiles placed on a 3D earth and instead using some technique which takes GIS data and procedurally generates the scenery.  However, you would need extremely detailed and accurate geographical data to know where to place, for example, an irregularly shaped grass field, a rectangular crop field, an irregularly shaped wooded area, a paved area, etc.  The you'd need to know the location and type of buildings or vegetation to place on top of those areas.  I'm not sure that GIS data with that level of detail exists, and certainly not for the entire world.

I think this is why MSFS chose to use satellite imagery and then imagery analysis to determine the types of objects to place upon the imagery in the sim.  It has drawbacks such as lack of seasonal changes and inaccurate buildings and vegetation, but I think that right now it's the best we can do.

Over time, I believe that the data will get better and better as well as cover more parts of the earth, so in the near future I can see a sim that uses entirely artificially generated scenery.

Dave

[MatthiasKNU 298]

6 hours ago, dave2013 said:

I'm not sure that GIS data with that level of detail exists, and certainly not for the entire world.

Yes, such data exist - partly on a regional basis, partly on a country basis, with huge data gaps also globally (OpenStreetMap).
The problem with these vector datasets is that (except for OSM, which cannot be used globally) all datasets have to be paid for - and are not cheap.

If we now use remote sensing data (i.e. mainly satellite systems), we have a spatial resolution of 10 m in the visible range (Sentinel-2). But to derive vegetation types we have to go into the multispectral range and here we are at 30 m/px (Sentinel-2 and Landsat-9 OLI) or 90 m/pixel (Landsat-8 TIRS). 

One step further, to be able to detect phenological state of vegetation, road surface, type of roof shingles, etc., we need hyperspectral data. Here we would be at the EnMap and PRISMA with 30 m/px. 

But to get field shapes, roads, etc., we need a hyperspectral data. Exactly, 10 m/pixel are often already too little. But as someone from DLR assured me, more than 30 m/pixel was simply not physically possible for EnMap.

All that is needed now is an algorithm that can process these petabytes and exabytes of data, and an engine that can display the results.

 

PS: By the way, what the vector landscape looks like based on 10 m/pixel data can be seen in the MSFS in regions where no aerial imagery is available, like here, south of Kisangani:

Edited September 11, 2023 by MatthiasKNU

[PavlinS 179]

14 hours ago, dave2013 said:

I like the idea of getting rid of the 2D texture/landclass tiles placed on a 3D earth and instead using some technique which takes GIS data and procedurally generates the scenery.

I am sincerely wondering how people will handle 3D data when this will get more performance heavy then 2D. There is a reason why we make 2D scenery looking like 3D. It more smarter than what you are suggesting. 

[MatthiasKNU 298]

1 hour ago, PavlinS said:

I am sincerely wondering how people will handle 3D data when this will get more performance heavy then 2D. There is a reason why we make 2D scenery looking like 3D. It more smarter than what you are suggesting. 

Everything displayed on a screen is necessarily 2D - but of course you try to add depth to it.
Of course, 3-dimensional data is inherently larger than 2-dimensional data.
But what is actually meant here is: Often it can be better performing to simply display 3 dimensional data than to calculate something that looks like 3D from 2 dimensional data.

[PavlinS 179]

19 minutes ago, MatthiasKNU said:

Often it can be better performing to simply display 3 dimensional data than to calculate something that looks like 3D from 2 dimensional data.

Let the processor do it's thing. So, you guys suggest we should all became 3D library creators for P3D. That means we work Blender or 3DSMax instead of using the libraries we already have. This is not just more labor for us, developers, but also increases the bottleneck on flight sim rendering. Imagine one photoreal scenery for China covered with handcrafted 3D models. This will be so performance heavy that even DLLS3 won't work on it properly. We all fly 3D models, you guys know how heavy a single PMDG 3D model hits performance. And you guys want us, developers, to learn how to make 3D models in 3D rendering programs to use in a scenery, with all them manual labor work that we need to do to create the autogen geographical data plus the download and conversion of the imagery. And all this trouble for a freeware distributed scenery? You guys, do you realize how labor intensive this will be? 

 

It is very easy to suggest new ways to create scenery, but this means a lot more ungrateful work for us, P3D developers. Step back and realize what you are guys suggesting. After all we are not super humans. 

And please, do not forget - there is bunch of versions of P3D. This means backward compatibility. This means all those new technology you suggest, we have to find a way to shove it in the backward compatibility so the scenery produced from us to be capable of using in all the P3D versions on the market. I am sincerely interested who among you will be the one to find a way to make 3D models capable for P3D v4 and v3. So we can actually use our creations in those simulators. After all we are not just an developers, we are also users of this platforms. For some of us this might be essential and we will want the data we created to be compatible with our flight simulators. 

I suggest those of you that suggest we use new technology to create scenery to think about all the users you are leaving behind to create all those new technology we are going to be unable to use. Because not all of us have only v6 on their pc's. 

Edited September 11, 2023 by PavlinS

[MatthiasKNU 298]

12 hours ago, PavlinS said:

So, you guys suggest we should all became 3D library creators for P3D. That means we work Blender or 3DSMax instead of using the libraries we already have.

No. Of course not - P3D is no capable of handling all of that.

As always, it's not so black and white - it's the mix that makes it. Therefore: Yes, in many places 3D objects in P3D are useful. But in many places, classic 2D elements are also what make it look good.

[d.tsakiris 418]

On 9/9/2023 at 10:50 AM, PavlinS said:

My personal opinion is that Asia should be the concentration of the P3D world

Oh yes, definitely...

[CO2Neutral 278]

On 9/8/2023 at 9:40 AM, MatthiasKNU said:

IF an engine would be able to reproduce in detail the light scattering, color rendering, all different surfaces, calculate all shadows (keyword raymarching reflections) in real time, take into account all parameters (weather, temperature, sun position, atmospheric thickness...)

Yes, the objective and much more realistic as all light is represented correctly.

On 9/8/2023 at 9:40 AM, MatthiasKNU said:

On a small scale, it may work great - but not yet on a global scale.

Not sure what you mean, but it is being done today and my guess will be available to PC crowd in the very near future.

On 9/10/2023 at 2:35 PM, Rogen said:

Yes of course a scenery fully decked out with objects is the ideal, but as you say that'll take years.

No, we have the technology, it's just costly at the moment and can be done without UV mapping.

We need to move away from the snapshot in time approach.

 

[MatthiasKNU 298]

31 minutes ago, CO2Neutral said:

Not sure what you mean, but it is being done today and my guess will be available to PC crowd in the very near future.

That would surprise me - we can't even capture all the processes that have an impact on "the way it looks" in REAL, so how are we supposed to be able to recreate that already on the computer?

[CO2Neutral 278]

14 minutes ago, MatthiasKNU said:

so how are we supposed to be able to recreate that already on the computer?

Welcome to AI 🙂

[PavlinS 179]

2 hours ago, CO2Neutral said:

Welcome to AI 🙂

Well, that's a good idea, can't deny, but I pref. to create AI that helps performance and at the same time to not lose the graphical quality of an scenery. Doing it as all 3D objects, sorry, this will be total waste of performance. I am watching some people streaming MSFS2020 on high end pc's and on landing they often have less than 20 fps with the PMDG 737. Imagine the level of hit on 777. So this is not the best idea by my opinion. Yet again I am saying. Better to find a middle way than destroy pc's with hard work on 3D models just for the sake of better looking scenery. 

[MatthiasKNU 298]

15 hours ago, CO2Neutral said:

Welcome to AI 🙂

As someone who has already trained AI systems and even programmed them himself: No, AI can't do that. 

[CO2Neutral 278]

5 hours ago, MatthiasKNU said:

No, AI can't do that.

I'll disagree, nVidia's providing some very nice tools one can leverage.

[CO2Neutral 278]

18 hours ago, PavlinS said:

So this is not the best idea by my opinion.

Fair enough, it's ok to disagree.