Overlapping UVs poll

PetrMaxa

I was having an argument with my co-worker other day about overlapping UVs. We both work in game industry.

We talked about AAA fps and his opition is that overlapping UVs on props are still relevant.  (He also likes to not have same texel desity in UVs )

I am trying to prove him he is wrong

Do you still use overlapping UVs on your models or not?

CrackRockSteady

it depends on the model and how it is being used but yes, overlapping UVs are still relevant.

If you're giving unique UV space to every single part of every single model you're probably being wasteful with your UV space.

Mark Dygert

Yes, when it's necessary.

Like Crack said, giving unique space to every UV shell can waste a lot of pixels, cause you to use big textures to get a decent texel density and all of that unique detail might not be required. If you give everything it's own unique space and are dogmatic about texel density you're going to have a lot of large textures that waste a lot of space. But of course that depends on the project and the art style.

Tiling textures and trim sheets exist for a reason. Both of those are stacking UVs. Imagine doing stairs with 100% unique space dedicated to each face, most of them look similar and can use the same texture, maybe offset a little to break up repetition. But again that depends on project and the art style.

If you're doing stairs like this, you can totally use tiling textures and trim sheets.



If you're doing stairs like this that have a ton of unique detail, you're probably going to generate unique textures, at least normal maps.


I also agree with him on texel density. It is a noble goal to strive for a consistent texel density. That is important but not all pieces are created equal, some pieces that just don't need the pixels and eat up a lot of space.

Maybe its a piece that needs to be capped to stop light from leaking into a room, but that cap isn't seen and doesn't need a lot of pixels.

If you go 100% unique unwrap, you end up forcing all of your other (important) pieces to be a little smaller, so a handful of unimportant polys can "technically" be the "correct" texel density, but it ends up hurting the overall piece. What good is checking off a technical rule if it made your stuff look like garbage?

Players don't care about your texel density adherence quotas.

  "This looks blurry and horrible but damn it if EVERY pixel isn't the EXACT same size! WELL DONE! 10/10"
- said no gamer ever

Your boss probably doesn't care either as long as something doesn't stands out egregiously. For players its visually for your boss it's probably visually and technically.

"We profiled your scene and you're way over budget, don't tell me you uniquely unwrapped all of this stuff? 3 4096's for those stairs? REALLY?"
-said the artists ex-boss

You learn why the rules exist, so you can ignore them, when they don't apply. If you just blindly follow rules and apply them to everything all the time, you can end up in a mess that isn't easy to clean up. 

Anchang-Style

You wanna uniquely wrap buildings? Looking at god of war, they had fantastic results with using pretty low res geo, putting tiling textures and then put a damage normal decal on the edges. Looks very good and keeps the res up. 

pior

Listen to your coworker.

SnowInChina

you're basically denying the existence of trim sheets in games Oo

defragger

You can use any trick if you understand its benefits and drawbacks.

RyRyB

Overlapping is still relevant as is min/max’ing UV shell density.

On weapons I would beg, borrow, and steal whatever I could to maximize density on shells close to the player’s view. 

musashidan

To NOT use overlapping uvs on geometry of the same mesh that you will never see simultaneously is just nonsensical and a complete waste of valuable texture space, and even if you can in the case of small details like nurnies you bet I'll be stacking them.

Think of the example of an FPS weapon. The very nature of the positioning on-screen means that the opposite side will never be seen anywhere near as often or prevalent(maybe during an animation) as the primary side that you're looking at all the time. Why would you give half of your precious texture space to something that will NEVER be noticed. And this applies to practically all objects in the real-world that it is physically impossible for us to look at from both side at the same time.

Joopson

musashidan said:

nurnies 

 What, pray tell, are nurnies?

[edit: Oh! They're the same as greebles. I should have known.]

musashidan

Joopson said:

musashidan said:

nurnies 

 What, pray tell, are nurnies?

[edit: Oh! They're the same as greebles. I should have known.]

Ask the bloke on the right...............

EarthQuake

musashidan said:

To NOT use overlapping uvs on geometry of the same mesh that you will never see simultaneously is just nonsensical and a complete waste of valuable texture space, and even if you can in the case of small details like nurnies you bet I'll be stacking them.

Think of the example of an FPS weapon. The very nature of the positioning on-screen means that the opposite side will never be seen anywhere near as often or prevalent(maybe during an animation) as the primary side that you're looking at all the time. Why would you give half of your precious texture space to something that will NEVER be noticed. And this applies to practically all objects in the real-world that it is physically impossible for us to look at from both side at the same time.

RE:FPV this is one of those things that makes sense conceptually but is applicable less often in reality. There's a number of disclaimers here:

1. If the weapon model will ever be seen in third person view, mirroring is often not appropriate, as mirroring artifacts (flipped text and such) will be visible from certain angles. You can make completely separate assets, but this is generally a waste of time.
2. Weapons need to be seen from the side and top, so mirroring down the center line is often a bad idea. Anything camera facing should not be split on the center line either.
3. Weapon designs are often asymmetrical and both sides can sometimes be seen in animations

Really, something like a FPS weapon is a good example of an asset that generally should be mapped unique, or at least mostly unique (areas behind the camera without specific handedness detail like butt stocks can sometimes get away with it). For something like a FPV weapon, you should be able to devote enough texture memory to it so that it looks nicely detailed, and doesn't have mirroring artifacts like it's 1999. You'll only ever have the one asset on screen, so memory can be managed in a sensible way (page textures in/out of vram when switching guns).

Though a FPV weapon asset is a good example of when non-uniform texel density can be very helpful. Generally areas further away from the camera or behind the camera should have less density, while focal points or anything like iron sights that the player can zoom in on should have more texel density.

Laughing_Bun

One other important note that I'm not sure was stated is making 1 really high quality trim sheet is much easier and faster than replicating all that high quality detail over a unique uv set. Less high quality stuff > lots of mediocre stuff

musashidan

EarthQuake said:

1. If the weapon model will ever be seen in third person view, mirroring is often not appropriate, as mirroring artifacts (flipped text and such) will be visible from certain angles. You can make completely separate assets, but this is generally a waste of time.
2. Weapons need to be seen from the side and top, so mirroring down the center line is often a bad idea. Anything camera facing should not be split on the center line either.
3. Weapon designs are often asymmetrical and both sides can sometimes be seen in animations

Really, something like a FPS weapon is a good example of an asset that generally should be mapped unique, or at least mostly unique (areas behind the camera without specific handedness detail like butt stocks can sometimes get away with it). For something like a FPV weapon, you should be able to devote enough texture memory to it so that it looks nicely detailed, and doesn't have mirroring artifacts like it's 1999. You'll only ever have the one asset on screen, so memory can be managed in a sensible way (page textures in/out of vram when switching guns).

OP specifically talked about FPS. As for flipped text, asymmetrical models, and centreline symmetry, these are obvious exceptions. I would never have centreline symmetry on an FPS weapon, but I would stack symmetrical islands from the sides that will never be seen(or even noticed during animation) together.

So these caveats are pretty much a given. This stuff is all based on common sense: what will be seen/noticed? What is important/prominent? What can I get away with without compromising results? This is all obviously going to be determined by what the player can see(TPV Vs FPV) and how close will it be seen, will it be animated? If so, can I get away with it based on the speed/angle of the animation? etc...