Will the Planetside TG2 developers ever consider using point cloud rendering in the future?
http://features.cgsociety.org/story_custom.php?story_id=5615&referer=rss
Sounds interesting. Although all the examples used are quite close shots which leaves the question if this is actually applicable for the massive outdoor scenes TG is used for. There's a lot more lighting of all kinds going on.
"Up" had extensively large scenes. This was used in "Up".
I didn't watch that movie but it seems to use a non-photorealistic render-style so realistic lighting (in a natural sense) probably wasn't a part of it.
Apart from that, what made me curious was the fact that 'Up' used this technique only in 90% of the shots. Since inter-diffuse reflection is basically needed in every artificial scene and point cloud lighting is a lot faster than radiosity I'm asking myself why they wouldn't always use it.
It's a very interesting optimization technique. Unfortunately I can't speak to whether it's applicable to TG2, but I also noticed the lack of any out the outdoor shots in Up as examples, and even if it was used in those cases, it's possible it was just in the closer parts (i.e. close to the camera). From the description of the technique, I'm not certain it could be applicable where e.g. you have a major diffuse GI light source (the sky for example) that is at relatively large distance from most surfaces.
- Oshyan
Per the realism - Christensen notes, "Point-based color bleeding has now been used in more than 30 movies. It seems that color bleeding has finally made it into the standard production pipeline in most of the large studios. This means that the movies area able to achieve new levels of realism."
I can't say I know about the huge distance scenes, but the lighting in "Up" was magnificent and made what looks simple 100 times better than it would be in any other rendering system in 3D animation.
Things like speed and realism caught my eye and then I wonder what's possible. Not trying to exhaust the Planetside crew. ;D
Quote from: Oshyan on April 16, 2010, 12:34:08 AM
It's a very interesting optimization technique. Unfortunately I can't speak to whether it's applicable to TG2, but I also noticed the lack of any out the outdoor shots in Up as examples, and even if it was used in those cases, it's possible it was just in the closer parts (i.e. close to the camera). From the description of the technique, I'm not certain it could be applicable where e.g. you have a major diffuse GI light source (the sky for example) that is at relatively large distance from most surfaces.
- Oshyan
The problem discussed in the article is that GI is barely used because it is too slow and that fill lighting setups are used to simulate color-bleeding and to obtain acceptable render time. I can imagine that the sky could give problems. However, the point of this article is that color-bleeding, a GI feature, is much faster and almost as accurate if not better when using point cloud rendering. That's why you don't see outdoor examples, because it doesn't really apply to it.
A possible scenario of TG2's rendering system could be:
The sky-system could be rendered with the raytracer. This already gives better results than the rasterizer, because you need lower amount of samples to get smooth results.
The terrain and objects could be rendered with point-cloud rendering. The article explicitely describes that point-cloud rendering is suitable for highly detailed displaced geometry. Also, distant geometry can be clustered efficiently to gain speed. We all know that if you render distant shots that the most distant part of the scene is the slowest by far.
In my opinion this is something which should definitly be looked into. 4 to 10x times faster, hard to ignore you'd say. Even if implementation in TG2 results in 2-3x reduction it is still a very interesting option.
Yesterday I saw some animated sequences of Vue in Clash of the Titans and they were so incredibly horrible ugly monsters :(
I bet the studio choose Vue over TG2 first because of Vue is known "better", but quite sure because of speed.
Optimizations like these could give a significant advantage?
Martin
I'd say that this might be even more worth looking into than GPU rendering?
From what I'm reading from this article, yes.
But, do I understand this like others (such as Matt) do? Not a chance. I was hoping to hear from him on this.
Quote from: Tangled-Universe on April 16, 2010, 09:37:01 AM
I'd say that this might be even more worth looking into than GPU rendering?
VUE is also easier to use right from the box; that's why I think many studios chose it. It must be frustrating to use, unless they made it more stable.
I'm not sure how often, since I don't work there, but I have understood from some who do that they take VUE and then do a lot of tweaking. So, they could just learn TG2 and make some tweaking toward more realistic, in my opinion.
We saw the same thing in Avatar. One of the lead artists used Lightwave and, because he was very good at it, he did well. It doesn't seem to be the tool as much as the artist. While I'm impressed with Rob Powers (http://www.robpowers.com/) and I was inspired to rethink LW, he's very skilled and so it wasn't LW that made the difference - it was Rob Powers.
Quote from: njeneb on April 16, 2010, 11:20:47 AM
VUE is also easier to use right from the box; that's why I think many studios chose it. It must be frustrating to use, unless they made it more stable.
Quote from: calico on April 16, 2010, 09:58:42 AM
From what I'm reading from this article, yes.
But, do I understand this like others (such as Matt) do? Not a chance. I was hoping to hear from him on this.
Indeed, not a chance. However, just by using common sense you would expect that it would be at least relevant to TG2, since it seems that point cloud rendering is especially suitable for lots of micropolygons. But perhaps the rasterize renderer is designed in such a way that it is difficult to integrate into the current renderer.
But we'll see. I too hope that Matt chimes in to give his thoughts about this.
"I can't say I know about the huge distance scenes, but the lighting in "Up" was magnificent and made what looks simple 100 times better than it would be in any other rendering system in 3D animation."
What are you basing that comment on? Even assuming (fairly, I'm sure ;D) that "100 times better" is mere hyperbole, I think it's rather presumptuous to conclude that the look of Up had much to do with this new rendering method. As the article states, what it does is make things faster, not look better.
As for the rest of the discussion, I think we're all drawing some very broad, sweeping conclusions based not only a short and non-technical article, but also on a very basic understanding of how TG2 itself works right now. I know when you say "you could use raytracing for the sky and point clouds for everything else" and "even a 2-3x reduction" it's really just a hope, but I think it's unrealistic to expect that kind of speedup, nor that implementation would necessarily be that simple.
In any case I'll try to have Matt weigh in on this. It's certainly interesting, I've just come to be very skeptical of any "magic bullet" optimization techniques, even if they come from production use with a major studio, because the things TG does are rather unique and seldom really tackled even by studios (most of the time they just use photo or hand painted matte backgrounds for this kind of stuff).
By the way, where was Vue used in Clash of the Titans? TG2 showed up in Imaginarium of Dr. Parnassus and Alice in Wonderland recently, so I don't feel too bad. ;) I would have been a bit more chagrined if the Vue fanboys had been right about Vue being used in Avatar (though I always knew it was just wishful thinking).
- Oshyan
Quote from: Oshyan on April 16, 2010, 03:58:28 PM
As for the rest of the discussion, I think we're all drawing some very broad, sweeping conclusions based not only a short and non-technical article, but also on a very basic understanding of how TG2 itself works right now. I know when you say "you could use raytracing for the sky and point clouds for everything else" and "even a 2-3x reduction" it's really just a hope, but I think it's unrealistic to expect that kind of speedup, nor that implementation would necessarily be that simple.
Of course it are mainly hopes :) You missed my point where I said that it is perhaps not possible to fit it into TG2's rendering system. I'm aware of it's unique-ness ;)
Quote
By the way, where was Vue used in Clash of the Titans? TG2 showed up in Imaginarium of Dr. Parnassus and Alice in Wonderland recently, so I don't feel too bad. I would have been a bit more chagrined if the Vue fanboys had been right about Vue being used in Avatar (though I always knew it was just wishful thinking).
It was used especially in the beginning when Olympus is shown first as kind of floating castles in the clouds. The clouds in the foreground were maybe not Vue but in the background they definitely were. You see the same Vue clouds in every render. Really couldn't believe my eyes they used them straight from preset.
A bit later some other similar shots from Olympus with huge Vue clouds.
@Oshyan - It was hyperbole. "Up" looks fantastic. This new rendering technique does make things look better. I can see that. That it's quicker by scads and scads of pixels per second is something I would hope could make TG2 better.
But, then again I have no idea. That's why I'm hoping to hear from THE MASTER. ;D Anyway, I bet he can tell us (me) a thing or two. :P
Quote from: Tangled-Universe on April 16, 2010, 09:32:11 AM
That's why you don't see outdoor examples, because it doesn't really apply to it.
Why wouldn't it? Color bleeding is basically color(and light) propagation via reflection. If you have a huge blue sky, objects in the scene reflect this blue light and it gets scattered further on to other objects.
An even better example are huge icy/snowy landscapes where bright sun light hits the surface and get's reflected up again onto the underside of the clouds.
Quote from: Rimmon on April 16, 2010, 04:45:52 PM
Why wouldn't it? Color bleeding is basically color(and light) propagation via reflection. If you have a huge blue sky, objects in the scene reflect this blue light and it gets scattered further on to other objects.
An even better example are huge icy/snowy landscapes where bright sun light hits the surface and get's reflected up again onto the underside of the clouds.
You didn't understand what I meant, or I did not say it clear enough.
I will simply quote myself again:
The problem discussed in the article is that GI is barely used because it is too slow and that fill lighting setups are used to simulate color-bleeding and to obtain acceptable render time. I can imagine that the sky could give problems. However, the point of this article is that color-bleeding, a GI feature, is much faster and almost as accurate if not better when using point cloud rendering. That's why you don't see outdoor examples, because it doesn't really apply to it.
With "apply to it" I mean apply to the examples and proof of principles the article tries to show.
Quote from: Tangled-Universe on April 16, 2010, 04:51:48 PM
With "apply to it" I mean apply to the examples and proof of principles the article tries to show.
Well, I guess I got that one wrong then.
Quote from: Tangled-Universe on April 16, 2010, 09:32:11 AM
In my opinion this is something which should definitly be looked into. 4 to 10x times faster, hard to ignore you'd say. Even if implementation in TG2 results in 2-3x reduction it is still a very interesting option.
That's 4 to 10x faster than a
purely ray traced GI solution. In Terragen the GI is never purely ray traced, at least not in current builds. A GI light cache is used instead, where most of the GI computation time is spent in the pre-pass. While this isn't the same as point cloud GI rendering, the speed increase is on the same order of magnitude at least, depending on your GI settings of course. There are advantages and disadvantages to any GI optimisation, but as is mentioned in the article, no single solution is a silver bullet.
Matt
How about GPU rendering? Is that something we could maybe see in the future of Terragen?
- Terje
Matt, thanks. Could you look at their code to see what we could take from it to do better? Oh, errr, this isn't Open Source, is it? Rats. :P
Matt's response tells me he's aware of the technique, as I would expect, and that he's already come up with his own clever approach that has similar goals (and may achieve similar speedups over a more traditional approach, particularly for our unique scene types). Which is to say I don't think this technique is likely very useful to TG2, based on his answer.
As for GPU rendering, it's certainly something we are looking closely at. The biggest concern right now is the fragmentation in the market. Nvidia has the edge on the actual software and development environment implementation, but ATI has the superior price/performance for current hardware. As far as market share I think it's split somewhat evenly for newer cards. DX10/11 and OpenGL 4 class cards are finally starting to bring convergence and a common set of features for GPGPU work, so I expect the cross-platform OpenCL option to mature faster now. Once OpenCL is in a better state and there is more widespread installation of DX 10/11 and/or OGL 4 cards, it will be a lot more appealing to do GPU accelerated rendering. In the meantime there are other CPU-oriented optimizations we still can make.
- Oshyan