Digital Foundry Article Technical Discussion Archive [2011]

[Neb]

They provide long distance visual scope combined with 24 hour lighting and smooth framerate, all on old console hardware. GTA additionally adds large amounts of mayhem to the mix and is able to somehow manage it all relatively smoothly. I'd personally add Assassins Creed 2 to the list as well. You can't bake your way out of visual issues if you have full 24 hour lighting, you have to make it all work. Getting stuff like shadows working at all times of day (both visually and computationally), getting textures to look nice at lunch time and under moonlight, making sure tone mapping gets the right look under bring sun and limited sun, etc, I mean it's a whole different ballpark of complexity. While the visual scope of a game like Halo Reach is real nice, it's fixed time of day makes it much less impressive, at least to me anyways.

Exactly, dynamic factor, not baking factor. So damn many exclusive games on consoles rely on TONS of baking and nice artwork for 'static' environnments. Sure it might look better but technically is a whole different playfield and most likely even inferior. It's like when I play ME2 and it gots lots of spots that look easily on par with best looking console games yet the game has lots of pre-baking, top artwork, lots highres and highly detailed non mapped textures (well most have mapping but still..) etc. But I wouldn't call it technically superior.

GTAIV impresses not just being so dynamic but also that shadows are updated per frame (AFAIK) with smooth shadow movement. HDR particles that have full world collision, POM effect for bullet holes, deformation, tire tracks, animation system etc etc. It's quite a big list.

 

[Laa-Yosh]

Yeah I know, figured I'd pick your brain a bit Wasn't sure if anyone was making more detailed use of these high res scanners for games as I'm out of the loop now.

Well, if you're working on games with a real-world setting, it makes sense to scan actual people, nothing beats that. Characters do take up a lot of the artwork, even if it's only the heads and naked bodies where you can really make use of the data. Celebrity likeness for sports games and movie tie-ins is another no-brainer.
LA Noire shows how much potential this approach has, especially when you include facial animation and have a lot of money.

But for anything else, or with lower budgets, you're not going to make much use of the approach. It doesn't even make sense to hire traditional sculptors and make maquettes of alien creatures or animals, that process is too slow and maybe even expensive as well.

One approach could be LIDAR scans, these are for very large areas - one movie VFX crew even scanned New York for a proper digital replacement, IIRC it was for "The day after tomorrow". But even that it's only good for reference, and when you have exact floor plans, blueprints etc. it's a lot more efficient and precise to use that.

 

[Laa-Yosh]

I was imagining a system that'd take a high density 3D model and 'vacuum form' a mesh around it.

There are several problems here. The vacuuming part exists, but prebuilt meshes are completely impossible for something like a car for many reasons. I'll get back to this part

There is research on automatic mesh generation and some apps implement it, but the results are always inferior even for organic stuff where messy construction could be tolerated to a point. Games also require efficiency, and animation has its specific needs. It's kinda hard for me to explain the attributes that make one mesh better than the other, even for people with modeling experience, it takes a lot of time and practice to learn this stuff. So please don't expect me to explain it to you

AI techniques would identify essential creases and map key points.

See, that is the very very hard part. Analyzing surfaces in 3D is probably very complex, maybe some programmer can help us out here... but I'd say even the identification of the key curves would be a problem. And since cars are highly reflective, even a small change can mess up the look, where shadow-light transitions are supposed to happen.

But surely there'd be a way to have a human brain select key data on the mesh and the computer to connect the dots and fill in the gaps.

That is what we do, but the amount of human decision and interactivity required is just too much to call it automatic. We don't manually place every polygon, but it's more complex than drawing a few curves or points either.

In the case of scanning cars for example, it's the same few basic forms in essence with variations on scales and postions.

Don't say that in front of any industrial designers if you want to survive the discussion! Seriously, cars (especially sports cars) are incredibly complex surfaces and works of art, there's a lot more going on that the average person probably couldn't tell. Even something as mundane as an Opel Astra has such quality to its surfaces and curves that it's very hard to model them, only the computing power of today allows the creation (and testing) of the body and its panels.

It's disappointing that 3D data capture and manipulation isn't progressing anything like 2D data has. I can use a digital photo or scan to get artwork into any document or application, but getting a 3D mesh needs someone to build it.

Well, I think that even a theoretical approach should reveal why the two aren't comparable and why 2D is actually not that advanced either. Most of the quality images are bitmap based and not vector based; those will always look inferior and can't get realistic beyond a limited level. The bitmap usually has very limited resolution, even at say 4Kx4K it's only 16 million elements and that's already enough for quite big pictures.

You could capture 3D in voxels and medical imaging is actually just that; but you would need several orders of magnitude of increase to account for the third dimension. 4Kx4Kx4K is 64 billion elements, and even with compression that's a lot of data. Now in a 3D world you can move up close to any element, and apply more magnification with narrow camera lenses, whereas you don't expect such levels of zoom from an image.
And 4K per dimension for a car would still only get you about 1mm precision, which is barely enough to roughly represent something like a car. It'd be about 10x better then lego bricks, but how well do those hold up at medium distances?

I'll try to show you a good selection of wireframe images to show you good mesh construction, maybe that'll help you see why computers can't really do it on their own yet.

 

[Barso]

I find RDR to be a graphical masterpiece.
Everytime I try something, it never fails me that it's been included. The whole game is a joy to play and watch. Such a shame it doesn't run 720p on PS3, I understand rockstar had this very problem with GTA4 and they still couldn't get it ro work by the time RDR was finished.
Although I applaud platinum games, they give bayonetta to sega to develop for PS3 and look what happened there, it's a real shame as it's the only reason I keep my xbox 360.
But they led development for vanquish on PS3 and we ended with a v-sync locked version on PS3 while the 360 version tears on 360. Same with criterion on burnout paradise.
So who really knows which system is stronger? They both seem to have their own different strengths and weakness.
I can understand the reason why exclusives look so good, because you can play to the systems strengths and hide its weakness and sony has without doubt the better first party games. I imagine they know this and I always hope they will make sure PS4 is a graphical powerhouse, they seem to be doing the same with PSP2.
Also they will have microsoft putting money into their pockets next-gen as the xbox360 successor will surely have a blu-ray drive.
I don't want to see sony go down the wii route.

 

[Shifty Geezer]

It's kinda hard for me to explain the attributes that make one mesh better than the other, even for people with modeling experience, it takes a lot of time and practice to learn this stuff. So please don't expect me to explain it to you

I can imagine! I've done a little hobbyist modelling and having to manage seams and UV coords, I can appreciate how mcuh more is needed for real work.

See, that is the very very hard part. Analyzing surfaces in 3D is probably very complex, maybe some programmer can help us out here... but I'd say even the identification of the key curves would be a problem. And since cars are highly reflective, even a small change can mess up the look, where shadow-light transitions are supposed to happen.

Converting 2D shapes to curves is possible. If a car were treated as a load of slices, like a CAT scan, a curve for each slice could be produced. Do that from all three directions and you'd have a wireframe. At least, that's what my brain does when imagining it!

You could capture 3D in voxels and medical imaging is actually just that; but you would need several orders of magnitude of increase to account for the third dimension....

I hadn't really considered that. The amount of raw data is going to be an insurmountable obstacle for a long time, until the 3D scanning method can actually sample the key contours and points.

Still, how's small-object scanning going? There was an optimstic movie clip for SOCOM I think where the props were being scanned in. Was that all just reference material for the artists to build up the actual in game meshes?

 

[Laa-Yosh]

Converting 2D shapes to curves is possible. If a car were treated as a load of slices, like a CAT scan, a curve for each slice could be produced. Do that from all three directions and you'd have a wireframe. At least, that's what my brain does when imagining it!

That wouldn't even work for like WWII airplanes, even those had some pretty three dimensional curves. Today's car designs are all about complex interplay between curves all around and the sharp lines would disappear in 2D slices.

I hadn't really considered that. The amount of raw data is going to be an insurmountable obstacle for a long time, until the 3D scanning method can actually sample the key contours and points.

Divide and conquer is the way to go. Scans capture the surface as a point cloud and it is a template for a more organized mesh that's ideal for animation and UV unfolding. Then you compare the simplified mesh and the source geometry and store the difference as a displacement map which is more efficient (only 2D information instead of 3D). This way the least efficient representation is only used during asset creation.

Still, how's small-object scanning going? There was an optimstic movie clip for SOCOM I think where the props were being scanned in. Was that all just reference material for the artists to build up the actual in game meshes?

I didn't see that video. But I guess they only used it for reference, a game mesh needs to be highly optimized.

You could probably run the high-res object through any normal mapper app out there but you'd need an extremely precise scanner for quality results. That'd cost you a few thousand dollars at least, if it can save you a few days of Zbrush or highres modeling work on a few dozen assets then you can earn its price back on a project.

But it's still a mixed bag, as it's way easier to use google for a few images, compared to getting your hand on something like an LWRC M6A2 for scanning. Literally everything is available on the internet, people have way too much free time to build websites and post photos
On the other hand, even in the US where people can purchase such weapons it's still unlikely that anyone on your dev team knows someone who owns the exact model you need, and then there aren't many games that have only a single assault rifle in them...

 

[Arwin]

It's disappointing that 3D data capture and manipulation isn't progressing anything like 2D data has. I can use a digital photo or scan to get artwork into any document or application, but getting a 3D mesh needs someone to build it.

I'm not sure if it was Kaz himself who said they had been testing with scanning, but found that the biggest problem there was that the results aren't optimised for the 3D engine requirements. The machine can't decide very well yet what parts it can leave out, what parts need more resolution and what parts less. I bet we'll see progress here, but obviously as requirements even differ per game/engine, it may take a while.

EDIT: "a game mesh needs to be highly optimized." <- yeah, what he said.

 

[JB9861]

It will be interesting to see the results of Infamous 2. If they achieve 720p at near 60fps (sandbox game), I wonder how well that point will hold up. The first Infamous was no slouch at 720p and around 45fps. The textures were still equal or better than the sandbox/open world games out now (and that was with a multiplat engine).

From what I've seen it would hold up well, despite the fact Infamous 2 is a generational leap over the first game in a lot of ways. The additions of SSAO, MLAA, and the fact their art is no longer at odds with their technology is doing wonders for the look of 2 nowadays. There's still a lot of culling though.

Now what would be truly interesting is Rockstar revising and tailoring a project to the PS3 as opposed to 360 like they've been doing. Which is why I've wondered what happened to Agent this last year...I'm hoping there's an update on its status this year.

 

[2real4tv]

Do any linear games have 24hr lighting? I would be nice to see Uncharted with that or even GOWIII. Did not play RDR but it looks good in vids and screens. GTAIV IQ is horrible, at least on consoles, so their was some compromises.

 

[nightshade]

Do any linear games have 24hr lighting?

Not that I remember, its going to be a waste anyway. Unless we are talking about linear games in a sandbox environment which is when I start to think of Mafia 2.

I would be nice to see Uncharted with that or even GOWIII. Did not play RDR but it looks good in vids and screens. GTAIV IQ is horrible, at least on consoles, so their was some compromises.

GTA4's bad IQ had a lot to do with the blur filter they had, have a look at BOGT on Xbox 360 where they removed the blur filter completely, it looks really nice.

 

[AlNom]

GT5 might count. Would be interested to know if it's only 24-hour cycle for the particular races that take advantage of that since I don't imagine it's a particularly good use of resources for races that are probably <6 hours, unless I'm missing something about the gameplay.

 

[Weaste]

IIn the case of scanning cars for example, it's the same few basic forms in essence with variations on scales and postions. A human user should be able to train an AI system to produce a uniform interpretation from any scanned car.

I've raised this point before, why when a game developer licences a car don't they simply get access to the original CAD models for that car to produce a polygon mesh from? Why mess around with car scanning? It's not as if the car industry didn't contribute towards computer graphics/modelling - Bezier was working in the field for Renault in the 60s.

 

[Laa-Yosh]

I'm not sure if it's always the case, but yeah, the original CAD models are way better than scanning. I'm sure Polyphony Digital has full access, not sure about the other, smaller studios though.

 

[Arwin]

I'm not sure if it's always the case, but yeah, the original CAD models are way better than scanning. I'm sure Polyphony Digital has full access, not sure about the other, smaller studios though.

Nope, PD has said that these are pretty much useless too!

As for the day night cycle, if you do an online race you can set the start-time, and the day night cycle will scale to the lenght of race you choose.

 

[LightHeaven]

Yes, I've been saying that for a long time - we should get Grandmaster on it.
It's amazing that games like RDR run better on 360, at higher visual fidelity despite having no HDD to cache to and running solely from DVD.




Well MS does not spend nearly as much money as Sony does on platform exclusives nor are their first parties as technically apt as Sony's.

However, you do have the occasional game like Alan Wake or even Halo Reach that can show what the platform can do. And who's to say that multiplat titles like Red Dead or Assassin's Creeds aren't as technically impressive as some PS3 exclusives

Perhaps Crytek's 360 exclusive will set a new bar for visuals on the system

I wouldn't boil all to this. They also go for different goals for their games. Ms pushes multiplayer gaming, almost all their games have to support local and online multiplayer, even on genres were no one else in industry does (Like Kameo, Crackdown or Fable). Given that they make so the visual hit are minimum with split screen that alone causes a huge hit on how good your game can look, and if you take in account the tight schedules and budget its really impressive what they do while keeping the framerate reasonably smooth.

Sony on the other side, has the market push that Ps3 is a bigger jump over the previous gen that 360, and that it is the most powerful console right now. So its their focus on the games they release, so they achieve higher visual fidelity at the cost of having less multiplayer features in their game.

If you look for games on 360 that has no multiplayer push, like Alan Wake, or in Ps3 games that focuses on multiplayer like Motorstorm, you'll see that they also follow that trend. Alan Wake pushes the system visually as much as they could, and Motostorm sacrificed graphics for multiplayer features.

Other than that, i'm not sure if that still the case this late in the generation, but a few years ago, people complained about sony keeping their best tools and hardware findings strictly to their 1st parties, while Ms since the beginning had their game fests with 2-3 editions around the world a year, where they would share new technologies and what they found to be best practices on their systems. Coupling that with the supposed more friendly dev kit and partners selling good technology like renderer engines, physics engines and etc, made 3rd parties multiplatform titles really shine this generation, being much closer to exclusives than the generations before.

I have to admit that they were impressive titles in their own way. But I didnt get the same wow factor I got from some PS3 exclusives. Alan Wake was the closest thing to Uncharted 2 but watching the two together you see lots of compromises done on Alan Wake like in character modeling, resolution, texture detail etc. If it wasnt for the scale and enemy numbers I wouldnt have considered Reach an impressive looking title. But that could be just me

But you will also find areas where alan wake pushed futher than U2, like having dinamic lights and shadows in larger scope areas with more objects and enemies in the scene when that occurs. I haven't played U2 throught to know if later in the game they managed that, but i've only seen that level of lighting in this game in confined areas, like caves or sewers. Alan Wake was also supposed to be an open worlded game, that ended up being linear, so their art had to pay the price of being open, even if the game ended up linear. Though i found AW really, really impressive while on the open, indoors, not so much.

As for reach, that's what is impressive with it, the scale and number of enemies is huge and that even then they managed to add a lot of finer details to their models. If you explore their assets while playing you'll find areas that even seem like a waste like modelled nails on structures that you wouldn't even be able to see with you stayed on the stage's course.

I've raised this point before, why when a game developer licences a car don't they simply get access to the original CAD models for that car to produce a polygon mesh from? Why mess around with car scanning? It's not as if the car industry didn't contribute towards computer graphics/modelling - Bezier was working in the field for Renault in the 60s.

I think because that could fall in your competitor hands and they would have basically your whole project to improve upon, including your whole aerodynamics.

 

[Laa-Yosh]

Nope, PD has said that these are pretty much useless too!

It's still good reference because it's better than a 2D blueprint. Getting it to a format that your average 3D application can read can be challenging though, as industrial design uses NURBS exclusively (it allows them very precise manufacturing, the computer data can drive all the tools).

 

[Arwin]

It's still good reference because it's better than a 2D blueprint. Getting it to a format that your average 3D application can read can be challenging though, as industrial design uses NURBS exclusively (it allows them very precise manufacturing, the computer data can drive all the tools).

Right, and then of course PD has to deal with cars for which such material never existed. So I'm guessing they'll be doing most of the cars the same way. But perhaps they do scan car toy-models and start from there - we do know at least that they use car toy-models as a start.

 

[ihamoitc2005]

Well MS does not spend nearly as much money as Sony does on platform exclusives nor are their first parties as technically apt as Sony's.

This is not true, my friend. MS spends more for much more for 1st party titles and even more for 3rd party titles with many paid exclusive arrangements and subsidy of development costs for new studios.

Just compare number of workers for Turn10 (300) with Polyphony Digital (140).

This is because Microsoft is not seeking profit but sales volume. If you look at financial statement, despite great game sales, millions of gold live accounts, and low hardware cost, and very high hardware margins on well selling accessories like Kinect, hard-drives, wifi, etc, xbox division still loses money.

Unlike Microsoft, which is the highest profit non-oil company in the world, Sony is forced to recover hardware loss with profit from game sales. They do this due to lower software development costs and no payout to 3rd party for exclusives.

It is actually silly to say that Microsoft has smaller budget for anything. They spend more on everything. For example, they are loosing $3 Billion a year on online services division. Only small portion of that cost is labor, servers, etc. Much of that is because online services is a channel to provide payout to publishers and others for promoting Microsoft products or supporting Microsoft market strategies.

Just look at this:

http://seekingalpha.com/article/246593-how-long-can-microsoft-s-burning-millions-continue

 

[TheWretched]

What first party studios does MS have left? They were all but disbanded in the recent years... they have some left, but they are just a shadow of what they used to be, especially since Bungie "left".

 

[Squilliam]

Just compare number of workers for Turn10 (300) with Polyphony Digital (140).

Where did you get the employee numbers from? Are they even current?

Anyway considering that we don't know how many games are under concurrent development at Turn 10 nor whether that employee headcount included contracters needed to finish a game or regular FTE numbers we cannot make any real comparison between them. I know that Turn 10 did mention they were interested in making other games aside from racing games however.