21 replies to this topic

[Ted_Ball]

This is just a quick question before I make some suggestions re reducing the poly count of all those trees. 

 

I might be on to something or I might not.

[frank70]

@Ted

 

I asked myself as well if every tree on some courses really is needed. In Schwarzwald i would suppose that you could leave maybe 20-30 % out. Trees in the fourth or fifth row cannot even be seen when you are playing (sight is blocked by other trees).

 

Probably we will see some "optimization by leaving unnecessary things out" in the future. The designers are relatively new to CF. Right now it's probably all about the beauty of the design (what is understandable). But down the road playability on as many rigs as possible will get more important. 

[Ted_Ball]

There's always got to be a compromise frank - a give and take. If you could make trees less of a burden on your system then you could have a resultant increase in another graphical something.

 

There is such a thing as 'occlusion'. If I'm right that means exactly what you are saying - something that doesn't have to be seen is not rendered. There will be people who will give us technical reasons why you can or can't have these things or not (to add to the confusion).

 

I look at a tree. I don't see every leaf, twig, branch, limb or even most of the trunk. But you will see every tutorial on tree building shows how to elongate the trunk, extrude limbs and generate every leaf even though it will have no effect on your round of golf or to be seen ffs.

 

I was going to illustrate by comparing real photos of courses we play against the PG render. In simple terms (and that's the only terms I have) a tree or the canopy is, in general, an ill-defined blob of a certain hue of green which has a cyber mass that is not even effected by sunlight. So why would you have your poor old computer draw every leaf within that blob? 

 

To blow my own trumpet, this is what I have been doing all my life. Not blowing trumpets - but using illusion as a way to convey the impression. Remember the Impressionists? Anyone? 

 

Anyway, more later.

[Dazmaniac]

Maybe so, as designers testing CF are often advised not to over plant high detail vegetation, especially in areas that are only really space fillers in areas surrounding the course that would be rarely visited when actually playing a round of golf. For this they are advised to use lower LOD versions of the trees, bushes, grasses etc.

[IanK]

The first course I created I was told that I had under planted. The second course I was told that I'd over planted!
Both played ok on my PC but I don't want to create a parkland or heathland course so that it's playable on the lowest spec PC and for it to look like the Sahara desert.

[Kablammo11]

I look at a tree. I don't see every leaf, twig, branch, limb or even most of the trunk. But you will see every tutorial on tree building shows how to elongate the trunk, extrude limbs and generate every leaf even though it will have no effect on your round of golf or to be seen ffs.

 

Actually, most tree models both in Speedtree or the Unity equivalent, do not come with individual leafs, but with branches and group of branches containing from 20 to 80 leaves fixed to the branches already. And these branches, twigs and leaves exist as textures with alpha levels for transparency.

 

I was going to illustrate by comparing real photos of courses we play against the PG render. In simple terms (and that's the only terms I have) a tree or the canopy is, in general, an ill-defined blob of a certain hue of green which has a cyber mass that is not even effected by sunlight. So why would you have your poor old computer draw every leaf within that blob? 

 

He needn't, the poor old thing, but there's a big BUT: In a 3D computer game, especially with following or ball tracking cams, the distant blob can easily come closer and move into the foreground, where it needs to be able to prove that it has individual leaves.

The tree distance setting determines the distance at which trees are not rendered as 3D objects anymore, but transition into very affordable billboards. At any moment on the course, the trees outside that distance are not affecting performance at all - to wit, the trees on the hills around Willow Heath are all billboards (the Links type of tree). And with the Speedtree LOD system (Level of Detail), trees in the middle distance are being dumbed down re: polygons and texture resolution, offering yet more performance economy.

 

To blow my own trumpet, this is what I have been doing all my life. Not blowing trumpets - but using illusion as a way to convey the impression. Remember the Impressionists? Anyone? 

 

Blowing your own trumpet about not blowing it. Good one! Give yourself a pat on your shoulder without patting your shoulder for that.

 

Anyway, more later.

 

Yes please, Ted. Minds need to be thoroughly boggled around here - and that is the most precious service that you have performed for this community during the past two years.

 

 

[Ted_Ball]

Yeah I can see the dilemma Daz. It's easier to just plant copies of trees than to take more time to make trees with a lower poly count (which have to be made). The dilemma being: take the easy option and incur the wrath of the public for a course that hits the frame rate or spend more time optimising trees. Maybe someone will produce a library of low poly trees for those out of the way places.

 

Anyway, more later.

[IanK]

Yeah I can see the dilemma Daz. It's easier to just plant copies of trees than to take more time to make trees with a lower poly count (which have to be made). The dilemma being: take the easy option and incur the wrath of the public for a course that hits the frame rate or spend more time optimising trees. Maybe someone will produce a library of low poly trees for those out of the way places.

Anyway, more later.

Mike provided some in the last package. The problem with them is that they appear a little too dark when used in the background. The speed trees seem to loose colour and definition and they fade into the distance but the low polygon trees stand out rather.

[Ted_Ball]

Is not SpeedTree a $ cost? But putting that aside - textures are what I'm moving towards as an alternative to individual components. Transparent textures are fine but I think you will agree that there is still an unnatural 'hand' of leaves when viewed.

 

Billboards are billboards and will always be thus based on their distance from the viewer. The loss of detail is very helpful system-wise. Not an issue. The ball travelling, unfortunately, through the foliage though will give the impression of travelling through a blob - granted - but if you've ever travelled at speed through foliage it will be a blur. You will have to take my word for it. Is it necessary to render this misfortune to a high level of detail?

 

Trumpetly speaking, I have been a photo-realistic artist in the past. I despised my inability to paint in the impressionist style - the mature style. Of course computer graphics rendered photo-real painting obsolete but strangely enough not impressionism. I nit-pick graphics in games. Driving from the cockpit in motor racing is a pet peeve. What do you see of your car when you drive in your car? It's only a peripheral blur (or impression of dashboard instruments and hands on wheel etc). Certainly not sharply detailed instruments - unless you need to refer to them. Same for golf games. (This is a level of game graphics that has been experimented with but never really been successful. VR will solve the problem in spite of itself.)

 

Anyway, more later. Nice talking to you K11.

[Mike Jones]

This is just a quick question before I make some suggestions re reducing the poly count of all those trees. 

 

I might be on to something or I might not.

 

You  can reduce the poly count of trees dynamically with the in game tree detail sliders. We already use lower poly count trees away from the lines of play. As a designer you can also raise and lower the individual poly counts of trees within the project.

 

The PP courses are generally very well optimised but the community courses will be a mixed bag.

[Ted_Ball]

That is wonderful Mike but it might reasonably be argued that 'Simplified Tree Detailing' will reduce the 'impression' of tree realism. There just might be a way to render trees as visually realistic while also reducing the poly count and, therefore, doing away with the slider compromise.

 

Later, there will be more, anyway.

[Ted_Ball]

 

Even though I didn't get an answer to my original question I'll assume that that is the case. I am pushing my argument by selective observation but learning a lot as I go.

 

 

Ed Fries, the co-founder of Xbox and the developer behind Halo likened the development of games to that of painting. Once painters were able to create realistic images, the medium began to stagnate and become boring because everyone began doing the same thing: recreating reality through painting. Likewise, now that games have the ability to create incredibly realistic experiences, the majority are doing just that. Painting managed to break free from this path thanks to movements like impressionism, which imposed artificial limitations on the artist. And Fries believes games may have to do the same thing.

 

"Pick some aspects to constrain and see what comes from that, see what creativity comes from that, see what kind of room for interpretation comes from that," he said. "The art world couldn't just jump to impressionism—they had to get everywhere else first and then they could go to impressionism. Maybe that's true for games. Maybe we're at that point now where we can do everything really well, and so now we have to ask ourselves what do we want to do to make things look different. Or are we going to just let all games blur together and look like the same kinds of characters, the same kinds of stories, the same kinds of visuals?"

 

And..... 

 

When it comes to poly count and poly reduction6 in games it all comes down to what the player will see in the scene and from what perspective the player will see it. Omernick (2004) explains the basics of poly reduction in his book Creating the ART of the GAME. He states that when reducing the poly count on a polygon model you should ask yourself what the player will see and delete unnecessary parts or faces of the 3D models that will not be seen in the game. You should also consider if all parts of the model are important or if they could be deleted in order to reduce the poly count. The third thing that Omernick mentions is to consider if the tessellation/subdivision level of the model is too high. He also states that a lot of details does not need to be modelled but can be implemented through texture maps.

 

 

The next step will have to be building a tree in Blender and through trial and error reduction of polygons while still giving the impression of a solid and dynamic tree. If you look out the window at a tree (assuming there is a tree in your neck of the woods (if you could call it 'woods')) then compare that to a typical PG tree then you will see they are quite similar and the rendition has been successful. Nevertheless, that rendition has a high poly count and, if there are a lot of trees on the course, the problem is increased.

 

I understand that a golfer can end up anywhere on the course but it seems perverse to have so much detail for a couple of mugs - bless them. I also understand a course architect doesn't want to spend much time optimising by addressing the out-of-the-way plantings. Some lateral thinking could be applied here. 

 

The areas of golf courses where there is more traffic will become more important to render at a higher LOD I would imagine - grass, waterfalls, etc making it necessary to reduce the impact of trees.

 

[DC#1]

Not sure how Unity or course forge work but I design for PGA 2000 and they have an optimization tool that you control the percentage of optimization by a slider bar thereby reducing the number of points and the size of the file. Granted it is only for textures and not trees. Is that what speedtree LOD does,basically optimizing at different levels? Does Unity or course forge have the ability to optimize textures, objects, etc?

[MimicPS]

One question I, as a non-designer but interested observer, is whether the LOD is applied universally across the object, or whether you can differentiate, or rather stratify, the LOD so that where you are most likely to spend enough time to observe (ie. at ground level when performing a shot) there is a high LOD and at canopy height where observation is moving and passing, you can lower the detail levels?

 

Would that actually require two objects to perform, or is it on a selective level? I apologise if that is a stupid question, but I have no experience at all of the tools in use.

 

I appreciate that to have the varied levels of detail would compromise the look of tree designs, but if a happy compromise could be found between two levels then it may be a way to squeeze the last few drops out of designs which were pushing the limits.

[Acrilix]

Not sure how Unity or course forge work but I design for PGA 2000 and they have an optimization tool that you control the percentage of optimization by a slider bar thereby reducing the number of points and the size of the file. Granted it is only for textures and not trees. Is that what speedtree LOD does,basically optimizing at different levels? Does Unity or course forge have the ability to optimize textures, objects, etc?

 

I've been fiddling about with the Speedtree modeller and it has built in optimising sliders to control the polys of trees you create.

[Kablammo11]

 

 

 

Ed Fries, the co-founder of Xbox and the developer behind Halo likened the development of games to that of painting. Once painters were able to create realistic images, the medium began to stagnate and become boring because everyone began doing the same thing: recreating reality through painting. Likewise, now that games have the ability to create incredibly realistic experiences, the majority are doing just that. Painting managed to break free from this path thanks to movements like impressionism, which imposed artificial limitations on the artist. And Fries believes games may have to do the same thing.

 

 

You're onto something there - though the story of Mr. Fries is incomplete. The reason why painting evolved in the 2nd half of the 19th century was, drumroll, the appearance of a very strong rival to painting: Daguerreotype, soon afterwards upgraded to photography. Before these, a lot of painters found work everywhere with lowly occupations such as local portraiture - depicting people, buildings, perhaps landscapes of their hometowns. That's the way their clients got pictures of things or people they cared about.

The triumphal advance of photography was a traumatic event and took away the livelihood of most painters within a few decades. And since realism was better served by photography, the smarter and more ambitious painters turned from craftsmen to artists, and occupied the territory of the abstract. With great success and beautiful results.

Same mechanism at work when the old-fashioned printers and typesetters were wiped out by DTP in the late 80s

 

Mr. Fries is... a sloppy researcher, at least, perhaps even bending the truth to suit his evil purposes. Painting did not evolve voluntarily because conventional pictures got boring, but because an outside agent drove it into new spheres. In other words, if you want better X-Box games, don't look to the current game developers to ever come up with the next step just like that, rather look out for a new force to appear and whip them into frenetic action.

 

But yes, well spotted, Ted: abstraction is the lovely and loveable daughter or realism. A dfficult, enticing and desirable lady to worship when angry outcries echo through these forums anytime somebody demands 3D grass or perfect water reflections witouth fps consequences, because he feels entitled and ends up feeling both insulted and offended when hearing the reasons why that is not advisable...

 

I do not believe, though, that a more performant and cleverer LOD system will offer a way forward. LOD, 3D meshes, textures and shaders are so last century, if you ask me. And it's the likes of Mr. Fries and his ilk that keep these outdated things alive to inundate their target demographic with painfully sub-mediocre fare, alas. An entirely new approach to all things 3D, built up from scratch with the possibilities of new machines in mind, is long overdue... Alas, the computer industry is stagnating and has failed to come up with any fundamental innovation in the last two decades. All they are able to offer are minuscule optimisations... the very thing your Mr. Fries lamented: Stagnation in boredom.

[DC#1]

Thanks Lez.

[J.H.Buchanan]

If ya make a course and they whoever has a problem with the trees to many or not enough save a copy of the exact course with no trees and then send this course to whoever and say "yeah you are right here is the course with no trees so you can put them in yourself where they need to be etc."  Ya may not get much of a peep out of them after that next time. 

[Quigs]

This is just a quick question before I make some suggestions re reducing the poly count of all those trees. 

 

I might be on to something or I might not.

 

Just my 2 cents. I don't really think of it that trees are the biggest resource hog as a blanket statement. It all comes down to poly's in a scene. Just as with any other game. Trees tend to be a big factor as there are so many of them. But...poorly designed object can have the same impact.

 

A quick example - I could build an awesome looking ball washer - modeling it to look exactly like those we see on our home courses. Nice and rounded, etc. Seems simple enough. Only to realize the poly cost of 18 of those placed around a course could have more impact than 100's of optimized trees. Same with rocks, other props, 3d grasses, houses, etc. And finally, there is the terrain, which has impact (flat vs mountains for example). 

 

The great thing about the tools out there, you will have the freedom to make your own design decisions. There are many techniques, LOD, etc, that help with providing a realistic canvas to work with. 

[Ted_Ball]

Point taken Quigs.

 

Another thing that confuses me is wind-blown trees. Is there some sort of optimisation of the animation that doesn't redraw the verts at each new point? To my mind a tree with all it's components moving with a wind effect should bring a computer to its knees. Obviously not. I suppose it is something like moving water.

 

Anyway, I'm still working on it. Blender has a Decimate modifier for curving trunks and branches that reduces polygons by 70% but retains the look. "It's the JPG compression of polygon meshes."