Procedural texturing for seasons/weather

[Murmur 3,941]

I thought more about the procedural/shader approach for depicting the various seasons/weather changes in scenery ( it was mentioned in another thread, more info and pics here: http://www.flightgear.org/forums/viewtopic.php?f=47&t=18603 ) and, IMHO, Laminar Research should start developing it ASAP and hence abandon the "classic" seasonal textures approach at once.

 

I can only see advantages of the procedural approach compared to the "classic" one, especially given the peculiar characteristics of X-Plane development:

 

1) procedural approach requires at most 1 single coder (and maybe 1 artist), while the classic approach requires a lot more "brute" work and hence a large workforce in terms of texture artists, if they want to deliver in short times;

 

2) procedural approach is by nature incremental. This is particularly in harmony with the X-Plane release model: they can release new shader effects as soon as they do a decent job (e.g. a shader making "autumn" colors), and then improve them and/or add new effects (e.g. "snow" or "spring"), release after release. Another advantage is that the shader would work on all scenery textures from the beginning; on the other hand, "classic" seasonal textures once made are "cast in stone" and cannot be modified/improved;

 

3) procedural approach can give much more realistic results: gradual accumulation of snow, gradual changing of colors (no instant switch between e.g. summer and autumn textures)

 

4) it is possible that procedural approach could be made to work natively even on photoreal scenery (BIG BIG PLUS!);

 

5) procedural approach takes no hard disk space at all, while adding seasonal textures would take at least various GB;

 

The only drawback I see is the overhead on the GPU due to the shader computations. Probably not much though?

 

Marco

[Geofa 108]

I agree!

[alpilotx 505]

Well, your analysis is not entirely correct. The procedural approach is - at least from the basic idea - a nice one. But the real problem lies ALWAYS - even here - with the part of the artistic work (which is very often vastly underestimated in many forum threads). The point is, that if you want a procedural approach, which looks really (and I mean rally - not just "more-or-less") good, you need a lot of more time invested per texture (yes, in the long run this should be still less than "redoing" each texture 4-5-times ... but I don't know by how much ... and hopefully don't need to rework all of them, when you discover some mistakes in your shaders approach). Because you need to "encode" the information, which pixels are affected and how by which shader (like: "what is foliage, which needs to become brown in autumn and disappear in winter", or "what amount of snow can be accumulated here" etc. etc.) ... You also need some very good foresight and some nice tool, which can simulate the shader effects to effectively create the textures. So, what I want to say is, that nobody should expect the "procedural" or "shader" approach to be a magic bullet, which removes most of the artistic work needed ... because if you want to make it "right", then it adds artistic work (but of course, in the long run, and if everything goes as "planned", you don't need to repeat it and can profit from "improved" shaders).

 

PS: and for this same reason you can't simply - and effectively - use it for any phototextures .... because without "good" info encoded in the textures about your foliage / snow accumulation etc. you can't get really good results (look at simulnights winter, and you understand what I mean --> it kind of works, but its not what I would call "good" in this sense).

[MSFS Ralf 149]

I'm an aircraft painter only and never worked on scenery texturer - but - reduce green, add red spots over trees and you have fall. Delete red spots and put white layer on top and you have snow textures. All add on layers set to multiplication, so the base texture stays visible. Not too complicated1

[Murmur 3,941]

Thank you Al, I was hoping on a reply from you since you're very knowleadgeable on these matters.

 

I still think that the procedural approach has short, mid and long-term advantages over the "classic" one.

 

The Flight Gear wiki does not mention any info encoded in the textures for "pixel color postprocessing" ( http://wiki.flightge...dural_Texturing ). For example, for snow coverage it only mentions terrain elevation and slope. Of course the procedural texturing will require lots of "rules" to get decent results, but I still think it should be significantly less work than drawing and editing thousands of textures by hand (x 4 times, at least!). And remember that many "summer" textures are still missing in the default scenery, meaning that Laminar is severely lacking manpower in the art-assets area. What hopes are there that they could create textures for the other 4 seasons? I think it's vastly more probable than a procedural approach could bring more results and sooner.

 

Also, all the other advantages over the "classic" approach that I listed remain.

 

PS: and for this same reason you can't simply - and effectively - use it for any phototextures .... because without "good" info encoded in the textures about your foliage / snow accumulation etc. you can't get really good results (look at simulnights winter, and you understand what I mean --> it kind of works, but its not what I would call "good" in this sense).

 

On this I don't agree. As I've already said, much better SimTiles than nothing at all, and probably the SimTiles algorithms/rules can be improved.

 

I understand you're a scenery developer and hence have a more "fundamentalist" approach :smile: (a little like me for flight models), but as Voltaire said, "Perfect is the enemy of good". You say SimTiles is not perfect... But is the cloud system in X-Plane 10 perfect? Is the weather system perfect? Is the ATC/AI perfect? i honestly don't think any of them is less distant from reality than SimTiles are.

 

Marco

[jcomm 5,930]

Murmur, great idea!

 

I believe some 3pd will certainly pick it ;-)... Who knows before Summer, and 20.30, we will get some seasons textures working in XP10 :-)

[alpilotx 505]

I'm an aircraft painter only and never worked on scenery texturer - but - reduce green, add red spots over trees and you have fall. Delete red spots and put white layer on top and you have snow textures. All add on layers set to multiplication, so the base texture stays visible. Not too complicated1

Yeah, that sounds like a piece of cake ... But there is not automatic process which can tell you in a textures where the tree pixels are. So, there you have the manual work, which has to happen (and doesn't falls from the sky). Then, you can't just remove that green ... somehow, you have to tell what should come instead of the green. Because even though many are thinking in terms of "lots of snow" when they talk about winter, but in reality - at least for example in most parts of Western Europe - you have a snow cover only 10% of the time ... The rest of winter is more soma dark, brownish stuff, with still a lot of pale, darker green tones (big meadows remain green over winter too ... etc. ... and heck, even conifers remain green ... but the tones change).

So, what I want to emphasize here is, that there is no magic bullet, and if you want to make it halfway decent, you need at least a pair of eyes and good hands to get this part of the job done (but before you stat, you need a concept, how your shaders would work at all ... ).

[alpilotx 505]

 

I still think that the procedural approach has short, mid and long-term advantages over the "classic" one.

Definitely! Here I am with you, and I think this is the way to go for the future (when ever that would happen - as I am not the one who decides). Especially because this allow for real "organic" changes in the landscape (as soon as everything is in place, and one has some nice shaders).

 

The Flight Gear wiki does not mention any info encoded in the textures for "pixel color postprocessing" ( http://wiki.flightge...dural_Texturing ). For example, for snow coverage it only mentions terrain elevation and slope. Of course the procedural texturing will require lots of "rules" to get decent results, but I still think it should be significantly less work than drawing and editing thousands of textures by hand (x 4 times, at least!).

 

Well, of course, there is a simple way, or a more advanced one. I - and I am not even an artist - would envision a little bit more than just use the slope information (which by the way is already used by quite a few shaders in XP10). The problem of an - almost - automatic approach is, that it can't account for many natural phenomenons, which make winter look like winter. For onw - as I noted in my previous post - you have to remember, that many time, in very large regions of the planet, winter is NOT white (in Western Europe its maybe 10% of the winter if we are lucky). Its some darkish, brownish stuff, and even with a lot of greens because of the meadows and coniferous trees etc. (so you can't just "automatically remove all green - only parts of it, which belong for example to coniferous trees ... but which is really hard to tell by just looking procedurally at an image). Then comes the next problem: how to cover (the hopefully already dark/brownish/green winter landscape) with snow, when it snows? For example, how would you - automatically - cover a forest with snow? The snow acuumulates very differently on the ground and the trees, and even after a long time, its not becoming just all white, but you can see some tree structures ... Whereas for example a similarly darkish field, would be covered much sooner and almost homogenously. So, the point is ... this is something you can hardly do automatically (maybe assist). Thats why I see here quite some artistic work ... a person, who can "encode" this informations (in what way ever - I don't have THE idea) in the textures (in an additional layer ... how ever that would work) ... or something like that.

 

Then we didn't even talk about all the other art assets! Most here might only see the ground textures, but there is a lot more, if you want to have a halfway plausible - oh, there we have that often discussed word again - scenery. What to do with the vegetation (it needs a completetly different "shading" approach, because its oriented differently than ground textures)? Again, you might need to encode, what would be leaves (which change color, or disappear) and what are branches (which remain in winter) ... and where snow can accumulate (on the branches, top-down ... but again, thats something you need to algorithmically tell the shader). Then comes all the other city autogen stuff .... with a lot of texture segments, where some are top-down, others are placed vertically ... and often represent very different materials. So, here again, someone (because algorithms might be easily fooled by our complex "structures") has to tell the shader, where (and how) snow can accumulate ....

 

So, thats why I say, that even though the overall idea is cool (and the way to go), it can't come like a magic bullet, and would really shine, if all the artwork is prepared for it (which is OK, because doing extra season textures would cost lots of work too ... so this preparation would definitely be worth it ... it just can't happen magically I am afraid).

 

And I remind you, that I am only telling you my point of view B) ... Where I am neither an artist (usually I don't do textures - beside the few in NZ Pro, or my Treelines/Farms scenery) nor am I the hacker of the rendering engine (thats Ben Supnik court) .... I am rather the data guy (imagine me like the one, responsible for the greenish symbol-mess which you see in the Matrix movies ... the artists put the beautiful colors on it, and the rendering engine makes it tick)

 

 

And remember that many "summer" textures are still missing in the default scenery, meaning that Laminar is severely lacking manpower in the art-assets area. What hopes are there that they could create textures for the other 4 seasons? I think it's vastly more probable than a procedural approach could bring more results and sooner.

Who do you tell that (with the "missing" textures ... but hey, at least I invested a lot of time, to have "old" textures be used in the Global Scenery 10 to they best until they get "replaced" step-by-step with new stuff ... which has happened in cooler climates already quite extensively).

 

Also, all the other advantages over the "classic" approach that I listed remain.

Which we don't need to repeat, because I fully agree here (and only point at the "hidden" difficulties, which are not necessary obvious the everybody - and can make this approach take much longer, than you might wish). I need to remind you, that I am neither the one, who will/can decide to go down this route (but at least I will for sure make my voice heard inside Laminar, when the right time comes for it - never ask for WHEN )

 

On this I don't agree. As I've already said, much better SimTiles than nothing at all, and probably the SimTiles algorithms/rules can be improved.

 

I understand you're a scenery developer and hence have a more "fundamentalist" approach :smile: (a little like me for flight models), but as Voltaire said, "Perfect is the enemy of good". You say SimTiles is not perfect... But is the cloud system in X-Plane 10 perfect? Is the weather system perfect? Is the ATC/AI perfect? i honestly don't think any of them is less distant from reality than SimTiles are.

I still say, for me I repeat: for me!) the SimTiles approach is not convincing (and why its so is easily understandable when you read about my thoughts above on what needs to be done for winter ... something, which can not really work in an automated manner). But this is definitely a question of taste (which I fully respect!). So, if you say, that everything is better than summer textures ... then its definitely a step forward! But for me, winter is a little bit more than putting some white (hmm, form the videos it often seems yellowish ... then I thin "don't eat yellow snow" :lol: ) over the landscape ... And I wont even stark discussing the "perfection level" of the rest.

 

But to sum it up: yes, I think this "procedural" approach to seasons might (quite likely) be the way to go, and no, it won't happen overnight.

[MSFS Ralf 149]

Edit: Posted at own thread, sorry for hijacking.

[Murmur 3,941]

Well we agree on a lot of things :smile: So at the end, I ask myself, should we realistically hope for a functional "season's shader" for X-Plane 11? I hope that's not too optimistic!

[Mountain Man 142]

The only drawback I see is the overhead on the GPU due to the shader computations. Probably not much though?

 

"Probably not much" is not a safe assumption at all. You referred to FlightGear which does use procedural textures, but on the FlightGear Wiki page it says:

 

"Computational cost: Especially on older graphics cards, multiple texture lookups combined with evaluating noise and other functions is significantly slower than just texturing the terrain with a single texture. Most optimization strategies used in other 3D games (such as filling the z-buffer early and discarding occluded pixels) do not work well for a flight simulator where most of the terrain is visible at any given time."

 

I also think you greatly underestimate the amount of work required to put such a system in place. alpilotx has already detailed what is necessary simply to prepare textures for procedural rendering, but then there's the matter of putting the code into the engine in the first place which could require a significant amount of work.

 

I think ultimately the solution will be a third-party developer will create some seasonal textures and a way of automatically loading them into X-Plane at the appropriate time (if I had the time and expertise, I'd do it myself; the way people are clamoring for seasonal textures, seems there's some money to be made).

[HiFlyer 9,359]

I would not expect to see this in XP first, or in fact anytime soon, if at all.

 

I would not, on the other hand, be surprised at all to see it in Outerra. And Viewterra had it long ago, though the engine was never fully developed.......

 

I would be one of the first ones clapping though if Xplane ever became capable of changing its seasons in real time as easily as it changes its time of day.

 

[MSFS Ralf 149]

This weather engine is a dream.

[Muskoka 104]

Thanks Devon, that does look great, amazing stuff.

 

p.s. Great quote at the bottom of you posts by the way, very much applies to some around here. ^_^

 

Glen