Pixel or Vertex more important, looking forward?

The extra work is significant; I might be boosting with the 3x multiplier, but it still doubles the time at least.

And the added value is, as has been stated here, only experienced by a very small percentage of the market, as demonstrated by video card sales - thus every studio lead and producer have to consider it a waste of human resources. Add in the ever increasing time for the creation of an average game's content, and the decision is made.

Edit - Or to make it simple and easy to understand: most sutdios are already having a hard time to create the art assets, without any fancy higher detailed versions. The whole industry has been in panic for a few years now, it's hard not to notice...

 

That is the same thing. If the high-detail model has N iterations, and you select N-1 iterations, you have a lower-detail mesh than the original high-detail model. I don't see why you want to argue over something like that. We both know exactly what I mean. And it is far from complete nonsense. It was completely uncalled for to say that.
I'd like to continue this discussion, but with someone with manners.

For the rest, it all depends on how good the artist is, and how good his tools are. None of that changes the point that some people want higher detail in the game geometry though. Whether or not certain game studios can or want to deliver that, doesn't change much about that.

 

Personally, I'm still not happy with Far Cry's textures. I just wish the textures could still be crystal clear when you get close to them.

 

Generally for subdiv high res model with N iteration, N-1 iteration is not a low polygon model. Not even its base can be consider a low polygon model. If it was true, developers would just made a simple sub 1000 polygon model and just sub div them and pray it will look good

 

The problem with any automatic LOD model generation methods is that there aren't any (AFAIK) that can produce effecient geometry. In general tesselation occurs on a full model scale which isn't necessary. An artist can generate a far superior mesh by hand tuning areas that need more polys (heads, shoulders, etc...).

Whatever you do you will need to hand tune the LOD models, which is impossible with the amount of assets in games today. The nice thing with textures is that you can very easily create high-res source art and then automatically scale it down to a variety of detail levels, the same isn't possible for meshes (yet!).

There's work being done in this area (for example progressive meshes in DX) but problems remain (I think mapping UV coordinates is a big one, I read something on this a while back). Anyway, this is my opinion on why developers don't include a higher poly option on models in games, they obviously have some reason or else they would be doing it, right?

 

Well if you just want to get ~1 pixel per poly I think automation will be able to do that better than any artist

 

You've summed it up perfectly well. Noone would build a 4000-polygon model by building a 1000-poly one and subdividing it... it'd be a total waste of polygons because every part of the model would be tesselated, instead of spreading out the detail wisely.

What would be possible, and should happen soon, is to put the tesselation into the GPU, because then the engine could scale the number of iterations depending on the GPU speed. And one could use displacement mapping to make up for the lack of geometry detail on the over-smoothed resulting surfaces. But until we have these two features accelerated in hardware, in a programable implementation, developers are stuck with triangle meshes... and getting the most out of them.

 

Yes, that is true. However, the current topic was about creating several versions of a model in order to be able to scale the poly count down for low-end systems. Thus the LOD models would be displayed in full scale, and this would show off any problems in their construction quite well.
Not to mention that the whole argument has not yet taken into account the level geometry, which is significant as well... LOD that too?

By the way one of the most common problems with aoutomatic optimizers is that they create long, thin triangles, which are bad for skinning and so on...

 

I must have completely different modeling tools from you guys. I can create control cages of any size, anywhere in the model, which will allow me to control the local tesselation quite well. I can also start with a box, model a human head out of it, and roll the iterations back down to a box.
Take a look at how this could be done: http://www.highend3d.com/global/tutorials/subdmodel/page1.3d

Besides, note that I wanted HIGHER poly meshes, not lower. So if you would just take the current lowpoly mesh and subdivide it, you'd already have something nice.

 

Well, if we assume that "polybump"-style modelling becomes common, then getting artists to create high-poly meshes will already happen. The problem comes in when you have them create more than one (relatively) low-poly mesh for the game. It wouldn't be trivial (you'd need to make certain the reduction in detail is done in an optimal way), and would require a good amount more data (you'd need a new set of normal maps).

 

Thank you - but I consider myself to be proficient enough in polygon modeling.

Something that has even its sharp edges rounded down, and loses a good amount of its volume due to the smoothing; and about a half of its polygon count is completely wasted. Remember the truformed weapons? And we're still not talking about animation/rigging related requirements... 3D content creation isn't that simple after all.

 

The point was to demonstrate what I mean. But I'm not sure if you took it that way. Not that I care.

We simply disagree. Perhaps I have better tools, or know how to use them better. Or perhaps we're not thinking of the same thing.
Can't you agree to disagree?

 

I already addressed that point in a previous post. You're good at going round in circles.

 

The point was to demonstrate that I have about 5 years of professional experience as a 3D artist, and I believe that I know the tools and technologies well enough to judge the validity of your ideas.

Or perhaps I know better.
I'm aware that I'm acting cocky here, but the conversation reached a point where I'm not willing to get into specifics again and again. My work stands behind my words, feel free to check my homepage.

 

You think you know better, what if I think I know better?
As I say, it comes down to agreeing to disagree. But if you prefer to act cocky and sling your resume around, that's up to you. I'm not interested. This is not a job interview.

 

What I sling around is my work - because as the saying goes, an image worths more than a thousand words. I simply don't expect anyone to accept my words just because I am ready to repeat them again and again. 'Nuff said, I've wasted enough time already.

 

i cant believe u guys are arguing about this, IF U WANT BETTER RESULTS NATURALLY IT TAKES MORE WORK.

edit : nm

 

I fail to see the relevance of your work in this particular discussion. So I fully agree, you've wasted plenty of time.

 

Generally control cages that's consider low polygon aren't optimal, neither for its high res model it generates, or for the low poly model that its going to be used for.

Yes, if all you do is subdivide, you can do that, but you won't get optimal result. We don't want your head to look like a box when you roll down the iteration.

 

Scali, Laa-Yosh -- stop or this thread is history. I don't care who's right or who's wrong, I just don't want either of you to keep with this tit-for-tat. Thank you.