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

Yes you can.

 

In gameplay You cant, in replays You can.

 

not when in 3rd person chase which is weird cause i think you can see them in the other views, especially first person view

model swap the instant the race actually starts

http://i37.tinypic.com/117fkma.gif

 

I dont understand why the two models use different backlights

 

my guess is it has to do with them probably outsourcing the assets and not checking for consistency

 

I had this discussion with someone before (maybe here, maybe elsewhere?) and while it's clear there is a model change, that is not the reason (I believe) that the view into your own car is impeded.

Firstly, there is likely some change in the lighting model for actual gameplay, with the sun position being the sole lightsource, whereas in the pre-race views of the car there is likely a lighting solution to best show off the car from the various angles, no matter where the sun is in the sky at that time. I'm sure you've noticed this that, no matter what track, the lighting in the car pre-race is the same.

Secondly, with the model change there is obviously a tint applied to the rear windscreen (as there is with most modern cars anyway), which in part will obscure the car interior.

Thirdly, and most importantly, in real life you often can't see car interiors through the rear view window. This is common when the sun is fairly high in the sky and there is bright sunlight... which is the standard lighting condition for Forza 3. You can test this yourself by walking a couple of meters behind a row of parked cars on a dull day (where you will be able to clearly see the interiors, and then on a very sunny day where the sun isn't directly above (maybe a couple of hours after midday) and you won't. The interior shadows of the car, along with the angle of the rear window and how it's interacting with the light creates that opacity. When I had this discussion before I took some (poor quality) pics on my cameraphone to show what I meant. If you want, I'll try and hunt them down.

Lastly, for KKRT to say that you can't see into the cars when racing is clearly false, because even in 3rd person mode you can look into the cars you are passing through the side windows to see the interiors.

 

There's too much window talk and not much polygon counting going on at the moment...

 

I'm more interested in how many the poly counts for Enslaved and Lords of shadows are for the characters. my guess leads me to believe Enslaved's is 20k+ and about 16k+ for lords of shadows. the arms and legs make Enslaved look a bit less but the faces have good articulation. with lords of shadows it seems the opposite, the faces are a bit stiff but the rest of their bodies seem good. (i think the main guy's weapons and upgrades show on the model.)

that's my opinion.

 

Which is completely ungrounded in anything, I'd add; starting with the assumption that each and every character's poly counts are the same within the game...

But Castlevania seems to have some detailed models so 16k is definitely too low. That's like, Gears of War 1, where the lead characters were noticeably angular in close up shots in the cinematics.
Also, Enslaved's main characters are actually using far more simple forms and shapes, especially the girl; but nevertheless, it is quite impossible to tell. I'd much rather wait for some first hand info from the developers, instead of just coming up with some random numbers.

 

Here is a comparison of the "Oddworld’s Stranger's Wrath" model:

old Xbox version (3k polys) VS PS3 version (20k polys) VS CGI version (529k polys):

http://www.thesixthaxis.com/2010/09/25/strangers-wrath-model-comparison/

 

That's interesting, the jump from 3k to 20k is much more dramatic than the jump from 20k to 529k. I would guess somewhere along the line the law of diminishing returns kicks in. Laa Yosh, I know every model is different but is there a general rule of thumb as to when diminishing returns really kicks in?

 

That jump also includes the addition of far better (higher res) normal maps with a more complex shader, which are meant to reproduce the surface shading detail of the highres model. You should also look at the big image here to see that there are in fact still some differences between the source model and the normal mapped one; pay particular attention to silhouettes. Close ups would make the jagged lowpoly edges even more evident on the game model.

As for diminishing returns, it's a complicated issue. The first rule is that hardware efficiency will break down with small or thin polygons, so you want to avoid them. But on the other hand the player character usually only takes up less than 10% of screen space, so it doesn't matter that much in this case. If you have a mass of enemies then it can be good to be efficient with them, but most games generally don't do that, and especially with shooters, enemies tend to be even smaller (thus using some LOD models anyway).
Cutscenes however are a different case, because there are close-ups - but performance is under full control, so efficiency isn't as important here either.

There are two usual uses for more and more polygons; one is if the character design has a lot of props and accessories to make it look more interesting. Like in a contemporary shooter, don't just have trousers and shirts but add ammo belts, weapon holsters, shawls, more pouches etc. etc. Gears of War takes this to extremes, but they're making sure that the details actually blend into the silhouette so very few of them have to be modeled and most are just baked into the normal map from the highres model. This Oddworld guy is a good mix of both, the straps on the arms and legs aren't really modeled but the little thingies on the hat are, and he has some extra pouches and a torn poncho. Also note that his head and torso is usually the most visible in cutscenes so it makes sense to spend the poly budget's majority there.
The other big use of geometry is to accomodate deformations for animation. Well articulated fingers, realistic elbows/knees/hips/shoulders, and in particular good looking facial animation all require a lot of vertices to add proper falloff to the deformations and in general have enough geometry to work with.

So, balancing just these few priorities is already complicated enough and makes setting any strict guidelines pretty much impossible. In the end, it has to look good and work at a stable framerate on the target platform, and your artists should spent the least possible amount of time creating it.

 

Posting this becouse seems future is ehading forsome really compelx scenery. Interesting techdemo with tesselation presentation and Nvidias next GPU. The last part of the video shows the real material, the tesselated Sci-Fi city.

http://www.youtube.com/watch?v=oCPMHIDpT88&feature=player_embedded


Interesting together with claims of Crysis being redesigned to take use of "GTX580" and Nvidias investment in Crysis dev/game.

 

The polygon count is impressive, the art design not so much

 

FM3 polycount

Extracted mesh from FM3 reveals actual poly count.

LOD0: 172753 polys (car selection)
LOD1: 45074 polys
LOD2: 21802 polys
LOD3: 13134 polys
LOD4: 6360 polys
LOD5: 2556 polys
Hi-res cockpit: 79,845 polys

source:
http://www.neogaf.com/forum/showthread.php?t=376652&page=178

 

Though one would wonder if tesselation is used with Xenos for menus/car showcasing. Anwyay it is pretty safe to assume the LOD1 is for racing whiel the LOD0 for menu/showcase scenes. Puts the poly counts up with NFS: Shift which is 40-60k for ingame race car LOD (no extra detailed LOD for menus present). I bet GT5 ends up around there to with similar LOD steps. Well see when it gets 'packed up'.

 

I would bet LOD0 will be closer to 200K. I would think the in-game cars (LOD1) will be at least 5 to 10K higher. Of course, that would be impressive with 16 cars.

 

Cool, what car was that for?

Its interesting to note that the best model is 170k which while high is nowhere in the region of 1 million polygons that Turn 10 was harping on about, not even half that.

For comparison are GT5s models still around 200k (like they were in Prologue) or have they got a bump?
And how many polygons in GT5s equivalent of LOD1?

 

Kazunori said, that some cars like WRC can have about 400k polys, and about LOD 1 we dont know, but close ups from car selection, replay and gameplay dont show much change in geometry in Prologue. In GT 5 one screen with jaguar from photo mode from game [AI car] and photo mode from car selection showed that engine had less details. Without proper comparison in photomode we cant judge exactly right now.

 

But one could wonder if tesselation is used for LOD0 in scenarios where LOD0 is used. Some BS shots of F3 cars shows tremendous polygon amount (wireframe mode).

Also if one assumes the LOD numbers are for the mesh in pic,
http://www.neogaf.com/forum/showpost.php?p=23635534&postcount=8892

then it is not a complete car as the rims and tires are missing. Seeing BS/photomode F3 shots shows very high polygon amount for rims and tires so as total it would certainly go above 200k.

Anyway the most important will be how visible LOD changes are. Rims are usually the first to go down in detail when wheels are spinning at certain speed and rim replaced with low detail rim + blur texture.

When the GT5 comes out people will do reverse engineering and pick the car meshes out and then one could see the LODs, polygon amounts and probably more (like in F3 .bin files).