"Yes, but how many polygons?" An artist blog entry with interesting numbers

Well, this one's clearly making the most out of a reasonable poly count:
Teuti_motion_03.jpg


Nice find!
It also looks like they apply an iteration of subdivision to the faces for the ingame cinematics, smoothing out the edges.
 
Well, this one's clearly making the most out of a reasonable poly count:
Teuti_motion_03.jpg


Nice find!
It also looks like they apply an iteration of subdivision to the faces for the ingame cinematics, smoothing out the edges.

That image I think shows a gameplay model. Thats the stance Snake has during gameplay when you aim with the knife.

Is the gray area on the face the subdivision you are talking about? What exactly does it do

edit:
snake_hirez.jpg


I wonder where this model is that taken from. It has crazy amounts of polygons. I suspect its either the in-game model in the last scene (which was much more detailed) or the model shown during the installation screen
 
Is the gray area on the face the subdivision you are talking about?

I think it's gray because it has a different shader and texture map than the rest of the suit. Makes sense to separate them, so that they can re-use the same head for all versions of the Snake character.

What exactly does it do

Subdivision? Inserts new vertices to the middle of the edges and in the middle of the polygon, for each polygon. This quadruples the polygon count of the model, and combined with some interpolation, smooths the new mesh down a bit as well. It can be repeated again and again to get a model that's perfectly smooth, but in practice the iteration count is obviously finite.
More advanced subdivision schemes are able to vary the iteration based on the view direction and distance to the camera.


I wonder where this model is that taken from. It has crazy amounts of polygons. I suspect its either the in-game model in the last scene (which was much more detailed) or the model shown during the installation screen

No, it's probably the highres model they're using to extract normal maps.
 
That image I think shows a gameplay model. Thats the stance Snake has during gameplay when you aim with the knife.

Is the gray area on the face the subdivision you are talking about? What exactly does it do

edit:

I wonder where this model is that taken from. It has crazy amounts of polygons. I suspect its either the in-game model in the last scene (which was much more detailed) or the model shown during the installation screen

It might be from the scene just before the last boss fight, involving the big oven. The model used in that scene, and the one used in the last fight seemed to be much higher detail to me.
 
Inserts new vertices to the middle of the edges and in the middle of the polygon, for each polygon. This quadruples the polygon count of the model, and combined with some interpolation, smooths the new mesh down a bit as well. It can be repeated again and again to get a model that's perfectly smooth, but in practice the iteration count is obviously finite.
More advanced subdivision schemes are able to vary the iteration based on the view direction and distance to the camera.

That basically explains everything that I would have eventually asked. Thanks.:yep2:
 
Well, not really, but what's going on here parallels NURBS and your line of thinking.

NURBS are mathematically defined surfaces that are broken into triangles, a step called tessellation. You can set whatever tessellation resolution you want, so could have a NURBS defined teapot rendered with 100 triangles for a rough approximation, or 10,000 for smooth curves.

Subdivision instead takes an existing triangular mesh and halves each of the existing triangles. This rounds out curves, but can't add detail. So for something like a sphere, you could have a 100 triangle sphere at a distance, and as it moves closer, subdivide it further into 400, 1600, 6400 triangle meshes, so up close to the camera it still looks like a smooth, round surface.

A NURBS football however would be defined as perfectly smooth curves and you could add more triangles more gradually to the rendered mesh as needed. NURBS are much more progressive.
 
Actually, the most commonly used scheme (Catmull-Clark) subdivides everything into quad polygons in the first step; triangles will become 3 quad polygons each, 5-sided polygons become 5 quads etc.

And neither subdivision nor NURBS can add detail on their own, the geometry has to be there already, wether by polygons of a control mesh or control vertices of a NURBS surface.
 
And neither subdivision nor NURBS can add detail on their own, the geometry has to be there already, wether by polygons of a control mesh or control vertices of a NURBS surface.
Yeah, but what I meant was a NURBS curve will inherently have more 'detail' (information), as...well, why else are you using a NURBS curve?! So a three point curve defined by standard vertices could be subdivided smoother, but not following any advanced contour. A three-point NURBS curve will define curve gradients which will provide placement data for all added vertices so they can follow a more detailed curve.
 
Yeah, but what I meant was a NURBS curve will inherently have more 'detail' (information), as...well, why else are you using a NURBS curve?! So a three point curve defined by standard vertices could be subdivided smoother, but not following any advanced contour. A three-point NURBS curve will define curve gradients which will provide placement data for all added vertices so they can follow a more detailed curve.

They're fundametally the same.
You thinking of the subdiv surface too much interms of the algorythm. Consider the mest in a sub div surface as a conrtol mesh rather than as verts in a polygon model.

A basic catmul clark surface is identical to a bicubic patch, and infact it's one way of evaluating a bicubic patches is to subdivide the control points.

What subdivs add is some degree of curvature at none regular tessalations although technically for most subdiv surfaces the degree of "smoothness" of the surface at none regular tessalations is "mathematically undefined".

What makes nurbs and subdivs less attractive for realtime rendering is the relatively high density of control points required for models of "normal stuff", high fidelity models usually require thousands of control points, meanin that even a single subdivision step puts you in 10's or 100's of thousands of faces. And even the base control mesh might not be simple enough for your Lowest LOD.
 
derr how many polys do killzone 2 characters have? i heard them say that a character has more polys than an entire level in killzone 1 but how many polys did killzone 1 levels have?
 
http://forum.beyond3d.com/showpost.php?p=1060441&postcount=82

Other than our local 3D artist's best guess, you're probably not going to find anything else unless GG has discussed it.

i refuse to believe that a an entire killzone 1 level had only 10-15k polygons. thats just silly. the poly count of killzone 2 characters is just noticeably higher than any other game out there. so i dont really believe that guy. and seriously the jak and daxter both have 10k polygons in jak2+. that statement says alot about this guy's credibility.
 
i refuse to believe that a an entire killzone 1 level had only 10-15k polygons. thats just silly. the poly count of killzone 2 characters is just noticeably higher than any other game out there. so i dont really believe that guy. and seriously the jak and daxter both have 10k polygons in jak2+. that statement says alot about this guy's credibility.

I don't think you've even considered the nature of high res modelling for creating normal maps. These are the Technology forums, so save the defensiveness for lesser places.

(And please, do reconsider your punctuation and spelling. Your posting quality has no place here. If you cannot discuss things in a polite matter, then consider this your last warning to get out. )
 
Dude...don't get so defensive. And don't be naive.

The devs could have meant the one most seen character throughout the game, who maybe has 20-25k polys. Regular ones will have half of that.
 
I don't think you've even considered the nature of high res modelling for creating normal maps. These are the Technology forums, so save the defensiveness for lesser places.

(And please, do reconsider your punctuation and spelling. Your posting quality has no place here. If you cannot discuss things in a polite matter, then consider this your last warning to get out. )

the models for normal maps have millions of polys. i dont think the devs would be talking about those models. they are talking about in game geometry. even if the latter sentence is wrong please dont imply that im stupid.
 
Back
Top