Are game graphics at a point of diminishing returns?
In that case, what do we mean by “diminishing returns”?
- Does it mean that an increase in the graphical budget of a game does not increase sales correspondingly?
- Or does it mean that, subjectively, we are at a point where things look “good enough”?
- Or are we talking negative returns? That increasing polygon count makes humans look worse due to the uncanny valley?
We could also look at a case study. Megan Fox and Stuart Compton have an article in the latest Game Developer Magazine about Ambient Occlusive Crease Shading.
(Disclaimer: I certainly respect the work done here, and I do not mean to pick on these authors. The stated goal is even of moving away from the “imitative” towards the “illustrative”, which I find good. And perhaps things will look different once they start moving.)
Nevertheless: How big a difference do you see between figure 5 with, and figure 6 without ambient occlusion? How much effort is it worth going from figure 6 to figure 5? Is this not diminishing returns?
(Illustration by Fox & Compton, from Game Developer Magazine, March 2008)
I do think that figure 5 looks more vivid and indeed is preferable to figure 6. Though this could of course just be an effect of colour balance or something; I have no idea what ambient occlusion is.
As to the bigger question: apparently, lots of people are happy to go out and buy Blue-ray players to replace their DVDs. I don’t understand why myself, but it’s a fact. People do want higher resolution, better colours, smoother movement etc. Perhaps there is no end to this – the general public will get used to whatever the state of the art is, and continuously refine their sensibilities so that they appreciate when something better comes along.
So in ten years time, people will say that _of course_ figure 5 looks better than figure 6, only people born in the 1900’s would be so primitive as to not see the difference.
I don’t think returns are diminishing, as long as you don’t stray too far from the current state of the art. Not for the cost of the hardware needed to display the graphics, because of Moore’s law. And not for the cost of producing all those models and animations either – not as long as we keep adopting new techniques for content generation. (Better motion capture, CI techniques for modelling movement and extending to new domains, evolved neural networks for controlling physics-based character models, etc.)
I stared at the pictures in this article for half an hour trying to discern why one was better than the other. The third picture in this set (not shown here) highlights the differences, but that didn’t help much.
I thought it was an April Fool’s day joke at first. I had to check the date on the magazine to be sure.
I’m sure the differences are readily noticed by experts in the field. But this leads back to the point of this post. Does the cost (in terms of both CPU and money) buy you much?
Unless the audience of your game is exclusively graphic effect programmers, I’m not sure it does.
As a graphic artist, even I am hard-pressed to tell the difference between the two figures. I don’t even know what I’m looking for…
I tried to detect the areas of difference using the old Photoshop technique, but unfortunately the hard JPEGing of the images makes it difficult to tell whether it is really picking up the differences, or just compression artifacts.
I can only speculate that the difference is more clearly visible in the original, hi-res images?
In this unscientific sample 33% of the audience does think that figure 5 looks better.
Yes, I would like to see the original images as well.
jesper,
i cannot help thinking of don quixote here, the windmills being the eternal quest for innovation and progression within a never-ending tech tree. can we, correspondingly, postulate a point of diminishing returns regarding
“gameplay”, too? e.g. when one has to be a trained ludologist to spot the differences between a well-balanced and a not-so-well-balanced variations of the same game?
Why are they showing ambient occlusion in a space where the light is dim.
And why is everyone getting so bent up about pictures? Last time I checked shaders need to be seen in action to be appreciated.
If you don’t really know what ambient occlusion is- it’s the concept of light hitting an object from other objects which creates an ambiance of light but it is simulated(in real time) however by doing just the opposite- creating shadows with polygons when they are facing each other with respect to angle and distance and having an ambient setting already defined in the material.
It creates a soft sky lit effect most of the time with soft shadows in creases/recesses and from the ground to the subject. Which is why it doesn’t make sense why they are using a dark setting to showcase it.
You can check out the example from nvidia for more on ambient occlusion:
http://http.download.nvidia.com/developer/SDK/Individual_Samples/featured_samples.html
Ah, the utility of ambient occlusion is much clearer in the Nvidia video. It does create a sense of presence for moving objects.
As I already said, perhaps I am a bit unfair to the authors, but this just isn’t the kind of step up that, say, texture filtering was.
I had a really hard time telling the difference as well.
Honestly, I do think we are slowly approaching a point of diminishing returns. The overall look of newer games is becoming quite impressive, so to improve we have to focus on the details. Where before, throwing more polygons made things look MUCH better, now things are so high in polygons, more becomes much less noticeable. It seems to be all about the effects now.
Personally, I enjoy a game that forgoes super effects and works towards a good art style. I absolutely love to look at Team Fortress 2, for example.
Last time I checked shaders need to be seen in action to be appreciated.
That’s my biggest issue with all these much-touted graphical prowesses and bullshots. Screenshots are pretty, but usually the animation isn’t keeping up with the still visual quality. Robotic NPCs are more often a bother to me than a lackluster texture. And most importantly, when I’m running through the halls of a dungeon in Oblivion trying to escape from a troll, keeping in track the way by which I came, looking at my compass at the bottom of the screen to orient myself, checking my health bar, with my sword taking about a third of the screen, how could I possibly care whether the wall polygons are jagged? The doors or other monsters I pass by could very well be the old models from Morrowind, and I wouldn’t much care. Maybe we haven’t reached a “diminishing return” yet, but it sure looks like a “no return” to me.
Reminds me a bit of the discussion about whether you can ‘really’ hear the difference between FLAC and mp3 at 320 kbps. Some insist that they can, others are sure that they can’t.
I know that I can tell the difference between a CRT at 75Hz and one at 80Hz, and make an effort never to sit in front of any CRT which updates slower than 76Hz. When I used to do a lot of drawing, I could tell a 91 degree angle from a 90 degree angle just by looking. Now, I don’t draw, and have lost that particular habit / ability.
Another reference point: Years ago, a ‘first world’ visitor brought a portable transistor radio set to a remote African tribe. We all know what kind of sound you get out of a 10cm mono speaker, and it was probably a shortwave signal, with plenty of (heheh) bells and whistles.
The tribespeople were amazed at the radio set, and were quite sure there must be tiny people inside talking and playing music. This was demonstrated to be false by opening the back of the radio. The visitor left the radio with them, and when he returned a few years later, he discovered that all of them could tell the difference between a transistor radio and a someone actually present. They were all amused to be reminded of their own previous inability to distinguish between the sounds.
So I think you can attune your senses to these kinds of fine differences, but at a certain point it’s difficult to distinguish between whether there are diminishing returns or whether you’re just developing Aspergers’ light.
Life is short, and if you have never tasted fresh onions, dried onions at least can give a valid impression, which is all that’s needed for a good game, in my opinion. (Although it is the juicyness rather than the authenticity that makes the fresh onion more appealing).
To be fair (before I launch into unfair tirade mode), if you’re showing off a new tech feature with art content that hasn’t been created to take advantage of it, the returns are always going to seem meager – see the raytracing renderer a guy wrote for the Quake 3 engine as further evidence. Technically quite impressive, but you have to squint to spot the benefits.
On the other hand, yes we are pretty much in seriously-diminishing-returns-land. The point isn’t whether people can spot the differences, it’s how much development effort is required to justify the feature. Per-pixel lighting and normal maps effectively doubled or tripled the time it takes to create an art asset, and while most gamers can see the benefit clearly, if you had to translate that inherently subjective benefit to an objective figure (in essence, the game industry’s standard practice of cost-benefit analysis) you’d probably be well shy of the 100-200% increase in workload. Nintendo’s strategy with the Wii is confirmation of this: why spend twice as much for a 10% benefit?
It is difficult to see, and we had this problem in the office as well. The fact is, you can’t tell the difference easily when they’re side by side. I suspect that perhaps the image processing we put on every image in the magazine may have further marginalized the difference… Regardless, when they’re laid on top of each other, the difference is much more striking. See here: http://www.shalinor.com/code.html
Mouse over the images to see the scene with and without ambient occlusion. To be fair, it is still rather subtle.
I would also disagree that increased polygon counts necessitate a trip to the uncanny valley, as you hint. It’s all about technique in art, after all. More pixels in digital photography show more flaws and more ‘reality’, for instance. I would also direct you to the in-game work of Takayoshi Sato: http://satoworks.com/
His stuff generally has a lot of emotional power and imperfection in its polys. I think people generally just don’t take the time to make that happen.
Branson, you are right – more polys do not lead to the uncanny valley. It is all about artistic style.
Ambient occlusion is an effect that sounds like a cheap radiosity, from what I understand. A simpler way to fake bouncing of light. I can’t tell how much it can add to games because of the crappy example but this is an important effect! You can take simple geometry and make it look amazing with some great lighting.
Game graphics aren’t anywhere near “good enough”. It’s still amazing to see what can be pushed out of hardware in titles like Gran Turismo 5, Metal Gear Solid 4, Crysis and the like. The poly counts are getting crazy high but lighting and shading seems to be the most important area now, to achieving realism. There’s physics too but that’s another story… ;)
Realism isn’t necessary to sell a game. Smash Bros. is approaching, or has passed, 2 million copies sold in North America and it’s not made with cutting edge technology. Its merits are mostly in its brilliant design.
What’s important is that if a game is pushing to be the most realistic, its design has to benefit from it. If you’re trying to make “The ultimate driving simulator”, you’re going to want to aim for as much realism you can cram into it. You wouldn’t necessarily want to push the graphical limits of technology to make something like Katamari Damacy, where the lack of realism adds to the fun of the game.
Have we reached a point of diminishing returns? These beautiful blockbuster games keep coming out so I suppose we haven’t reached that point yet. When all these companies start going out of business that’ll be the sign to slow down, if most companies haven’t figured it out by now.
As Dominic mentioned, one of the biggest problems with realism in games is animation! Games look great but their cinematics and gameplay have terrible “game animation”. That almost always kills things for me. I haven’t seen much of GTA4 but I hope they went all out on the game animation.
Sure there are diminishing returns.. but the example is a bad one. First you took a terrible picture out of many possible ones. Then dynamic ambient occlusion has now been added to a lot of new games because of its bang per buck. It’s very cheap in term of development and doesn’t require a lot of authoring.
Now good animation requires authoring.. And why do you suffer with bad animation and try to make a distinction between animation and other graphics related topics ? Why don’t you agree that cartoony and unrealistic animation is good enough ? (rhetorical question) Well simply because it doesn’t help you relate. You know it doesn’t behave like that in the real physical world.
See : even movies highly stylized like Pixar movies are spending millions in research to improve rendering quality, light interaction, materials, and yes animation. Why do they do that ? Because it helps you relate better to what’s on the screen. You don’t think “oh that’s a horrible rendering” but instead of maybe they have some ground in reality.
Have you seen Little Big Planet. Big cartoony game but with realistic cues that make you think, maybe those characters are really done out of fabric and so on.
Ambient occlusion (and more generally realistic shadows and light interaction) participates to this. If well done, it helps see depth and how objects are related to each other (is that box hovering or sitting on that desk ?). Also if your art is not contrasted, it makes it pop up more.
@greg As I said earlier in the comments, the technique is more convincing in videos. I also agree that it makes objects feel more present on screen.
But I also think that a lot has changed in the five years since I posted this originally: Today we have much more of a consensus that diminishing returns is a real issue.
I think it all depends what you are trying to achieve – the vast majority of games played have unrealistic animations and so on, but often stylized in a particular way that people seem to enjoy.
Also, I do think that games and other art forms are also attractive because they are in many ways unrealistic and artificial.
I am not saying that you shouldn’t invest in improved gfx tech, just that it doesn’t seem to be paying off as much as it used to.
I personally think that we’re still at the bottom of the mountain.
While it can seem scary there’s vast untapped potential and regular advances in technology are definitely helping.
The only thing annoying to me, is that it seems to not progress “fast” enough ;).