Why aren't graphics done as vectors?

[Bad Jim]

All of which comes back to the initial point that it is more computationally expensive than other graphics methods.

The extra expense is justified. Up close, bitmaps look terrible and it's worth a little extra computation in return for not looking shit. Far away, bitmaps look fine, so we just switch to bitmap mipmaps and suffer no wasted computation.

 

[ThriKreen]

The extra expense is justified. Up close, bitmaps look terrible and it's worth a little extra computation in return for not looking shit. Far away, bitmaps look fine, so we just switch to bitmap mipmaps and suffer no wasted computation.

Vectors are much more expensive when it comes to CPU power to justify the cost if you're trying to use them to replace textures. Even a basic square could get rendered as 2 triangles, so your 5000 triangle model could jump to 50,000 or 500,000 triangles.

And that doesn't even factor in the creation of it, having to train artists to paint in vectors than as a regular bitmap style image.

 

[Eclectic Dreck]

All of which comes back to the initial point that it is more computationally expensive than other graphics methods.

The extra expense is justified. Up close, bitmaps look terrible and it's worth a little extra computation in return for not looking shit. Far away, bitmaps look fine, so we just switch to bitmap mipmaps and suffer no wasted computation.

Considering that the same can be achieved without computational expense by simply having multiple levels of detail for textures along with the capacity to buffer and load them quickly and reliably, vector graphics again fail to impress. Until such time that the computational expense problem is solved or rendered moot (when triangle density equals pixel density), the trade off simply isn't a useful one.

Especially when your dealing with either ancient hardware already being flogged to an inch of it's life using simpler technology or, on the PC side, such an abundance of memory that having dozens of levels of detail for any given texture is perfectly possible.

Moreover, the problem you seek to solve isn't a particularly common one in most games. It isn't often that we are asked to closely examine a texture.

 

[Hagi]

I don't see it happening even in the close future.

Vector textures have insane file-sizes, for that same size you could have many bitmap textures. A few small ones for long distance viewing and several huge ones for up-close viewing.

The amount of resources expanded on making the vector graphics work, both processing power and memory, are probably better spend on those multiple bitmap textures and switching them out as needed (which is possible although resource intensive, but since we're comparing it to something that's even more resource intensive it's entirely viable).

 

[veloper]

Simple shapes that are very suitable for vector graphics, such as big logos, are often modelled as polygons already.
This is nothing new.

 

[Bad Jim]

Vectors are much more expensive when it comes to CPU power to justify the cost if you're trying to use them to replace textures. Even a basic square could get rendered as 2 triangles, so your 5000 triangle model could jump to 50,000 or 500,000 triangles.

You can't compare basic 2D triangles with 3D triangles that need their vertices transformed, their normals transformed and a projection matrix calculated before you can begin texturing.

Considering that the same can be achieved without computational expense by simply having multiple levels of detail for textures along with the capacity to buffer and load them quickly and reliably, vector graphics again fail to impress. Until such time that the computational expense problem is solved or rendered moot (when triangle density equals pixel density), the trade off simply isn't a useful one.

But you can't just keep loading bigger textures because there's a limit to how large the game can be. It has to fit on a few discs or be readily downloadable. While vectors, though not adding extra detail, succeed in not looking shit when you zoom in.

Especially when your dealing with either ancient hardware already being flogged to an inch of it's life using simpler technology or, on the PC side, such an abundance of memory that having dozens of levels of detail for any given texture is perfectly possible.

The PC has plenty of memory but there is, once again, a limit to how much can be delivered on discs or downloaded from the internet. And I'm still not convinced that even the aging consoles would have that much trouble with it. They run at 30 fps because competition forces AAA devs to have as much detail as possible without the fps getting worse. Even if they became a million times faster, devs would just add more detail and effects until they slowed back down to 30 fps. Just because they run at 30 fps and are inferior to PCs doesn't make them unable to handle a simple pixel shader.

Moreover, the problem you seek to solve isn't a particularly common one in most games. It isn't often that we are asked to closely examine a texture.

It's likely that developers specifically avoid giving us a reason to look closely at random things because they know how bad they will look. You might have noticed that games hardly have any secrets anymore. You might have noticed how linear many games are. Perhaps if you could get closer to random objects without the textures looking bad, they wouldn't be so shy.

 

[sapphireofthesea]

Vectors are much more expensive when it comes to CPU power to justify the cost if you're trying to use them to replace textures. Even a basic square could get rendered as 2 triangles, so your 5000 triangle model could jump to 50,000 or 500,000 triangles.

You can't compare basic 2D triangles with 3D triangles that need their vertices transformed, their normals transformed and a projection matrix calculated before you can begin texturing.

Considering that the same can be achieved without computational expense by simply having multiple levels of detail for textures along with the capacity to buffer and load them quickly and reliably, vector graphics again fail to impress. Until such time that the computational expense problem is solved or rendered moot (when triangle density equals pixel density), the trade off simply isn't a useful one.

But you can't just keep loading bigger textures because there's a limit to how large the game can be. It has to fit on a few discs or be readily downloadable. While vectors, though not adding extra detail, succeed in not looking shit when you zoom in.

Especially when your dealing with either ancient hardware already being flogged to an inch of it's life using simpler technology or, on the PC side, such an abundance of memory that having dozens of levels of detail for any given texture is perfectly possible.

The PC has plenty of memory but there is, once again, a limit to how much can be delivered on discs or downloaded from the internet. And I'm still not convinced that even the aging consoles would have that much trouble with it. They run at 30 fps because competition forces AAA devs to have as much detail as possible without the fps getting worse. Even if they became a million times faster, devs would just add more detail and effects until they slowed back down to 30 fps. Just because they run at 30 fps and are inferior to PCs doesn't make them unable to handle a simple pixel shader.

Moreover, the problem you seek to solve isn't a particularly common one in most games. It isn't often that we are asked to closely examine a texture.

It's likely that developers specifically avoid giving us a reason to look closely at random things because they know how bad they will look. You might have noticed that games hardly have any secrets anymore. You might have noticed how linear many games are. Perhaps if you could get closer to random objects without the textures looking bad, they wouldn't be so shy.

Sorry, I just have to call you out on the last point. IF a developer wanted to use close-ups in a meaningful way they wouldn't re-invent the wheel to test it out, they would start with limited scope cheats (such as the multiple layers one) so that the general area you are meant to look closely at works. IF you handed the secret to vector development to the AAA companies today you would just get another CoD in vector, assuming it ends up being the cheapest options (which means good translation of out of house rasters into vectors)
I use both formats in GIS work and can tell you, vectorizing a raster map is a day long wait when that raster map covers ~100km by 100km area. A poorly done vector looks ALOT worse zoomed in than any raster, because of join issues and gaps, and industry goes cheapest option, you give them vectors and you won't see any increase on zooming, as it would be cheaper anyway to keep it accurate to a certain zoom and no more.
The process itself needs development and as other have said, it could offer some advantages, but does it offer enough to make it worth re-equiping the entire industry and will it take a form which doesn't just end up being make a raster and then vectorize, in which case you are double paying and would reserve the vectors for specifics anyway, and it wouldn't make zooming in more common place.

 

[Bob_F_It]

Considering that the same can be achieved without computational expense by simply having multiple levels of detail for textures along with the capacity to buffer and load them quickly and reliably, vector graphics again fail to impress. Until such time that the computational expense problem is solved or rendered moot (when triangle density equals pixel density), the trade off simply isn't a useful one.

But you can't just keep loading bigger textures because there's a limit to how large the game can be. It has to fit on a few discs or be readily downloadable. While vectors, though not adding extra detail, succeed in not looking shit when you zoom in.

But push that limit on file sizes much faster for the same level of detail.
Also, they eventually look shit. If you zoom in enough to say, the head of your example image, it eventually ends up as straight lines. So you're only postponing the issue to Bézier, field, or locus based graphics. All of which ruin frames-per-second even faster.

 

[ThriKreen]

You can't compare basic 2D triangles with 3D triangles that need their vertices transformed, their normals transformed and a projection matrix calculated before you can begin texturing.

I should have said more CPU/GPU intensive, but the point was you still have to apply the vector graphics from 2D into the 3D surface, either as is in some manner of projection, or rastering it into a 2D bitmap to be applied (in which case, why not just keep it as an easier to manipulate bitmap and rely on the mipmaps to handle your detail as opposed to rasterizing it on the fly). Either way, still requires manipulating the vector graphics so yes, you can compare it.

Probably easier to just model out the detail to get the same effect, which still slows the processing, even more if you start doing skin deformation for animations, so we're back to square one - it's slower.

 

[kenadian]

Because rasters take up less data and are more practical. That's really it.

 

[Bad Jim]

Sorry, I just have to call you out on the last point. IF a developer wanted to use close-ups in a meaningful way they wouldn't re-invent the wheel to test it out, they would start with limited scope cheats (such as the multiple layers one) so that the general area you are meant to look closely at works. IF you handed the secret to vector development to the AAA companies today you would just get another CoD in vector, assuming it ends up being the cheapest options (which means good translation of out of house rasters into vectors)
I use both formats in GIS work and can tell you, vectorizing a raster map is a day long wait when that raster map covers ~100km by 100km area. A poorly done vector looks ALOT worse zoomed in than any raster, because of join issues and gaps, and industry goes cheapest option, you give them vectors and you won't see any increase on zooming, as it would be cheaper anyway to keep it accurate to a certain zoom and no more.
The process itself needs development and as other have said, it could offer some advantages, but does it offer enough to make it worth re-equiping the entire industry and will it take a form which doesn't just end up being make a raster and then vectorize, in which case you are double paying and would reserve the vectors for specifics anyway, and it wouldn't make zooming in more common place.

To be honest, I wasn't really thinking of vectorising everything. What I had in mind was doing the broad strokes of the texture in vectors, by hand, but filled in with raster textures. The goal being to make features like writing, different materials etc well defined. I do agree that some sorts of texture, such as rock, just can't be effectively vectorised and detail textures are a better idea. But detail textures won't make a large blurry boundary look sharp, while vectors are perfect for this.

 

[evilneko]

Here's a game with entirely vector graphics [http://en.wikipedia.org/wiki/Star_Wars_(1983_video_game)].... back in 1983.

 

[wulf3n]

Not only computationally expensive (like everyone has already said) but also they don't look as good as regular bitmap images.

Sure in games you're not painting works of art, but pixel based allows a lot more freedom when it comes to blending and shading.

 

[Scarim Coral]

But wouldn't that take a HUGE chuck of the memory space?

Granted the last time I've used vector was a few years ago but still wouldn't the memory be huge for the game to have vector graphic?

 

[The Heik]

In this day and age people are obsessed with graphics, especially when they see the pixels. So why don't graphics designers get around this by using vector based art programs?

That explains what they are.
How it works is that the art is made out of lines, this allows you to zoom in on images without losing quality. This solves the problem of snipers using their snipers rifles to zoom into shoot someone and then see the ugly textures that are very low resolution.

They don't use it because the mathematical equations bumfuck standard processors. When something is rasterized, it's constant. When you make it big or small, you are not changing the file's size, merely how you view it which doesn't take much time or energy. Vectorized images and textures however must recalculate the image each time it is re-sized. Done with a singular static image, most computers can handle that without breaking a sweat, but you suddenly force them to process not one but most likely dozens of moving vectorized items and even HAL9000 would melt trying to compute all those equations in real time.

Maybe at some point we will use vector images, but right now we don't have anywhere near the firepower available to the mass market to make it a viable choice.

 

[Tanakh]

-snip-

Lolz. We finally agree on something ThriKreen! Must be a christmas miracle.

Happy holidays mate.

-Snip-

Dude, we have two options. Either try to implement a new way to use vector images using more GPU time on rendering and slowing things down, or just use high def textures (like we do) on the important stuff which only taxes the memory (which is dirt cheap nowdays anyway, even on graphic cards).

Why would anyone ever go for the hard, complicated and inefficient way?

If anything, things look bad ATM on a sniper zoom because our current console generation is ancient and in the ol' days when they were designed memory was way more expensive. Go and snipe in BF 3 on a good PC with high res textures and I doubt you can complain.