1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Vertex Precision / FM3 conversion LOD0

Discussion in 'Racer Physics and Technical' started by QuadCoreMax, Oct 18, 2010.

  1. hey everyone,

    Been lately playing with FM3 cars & successfully converted some of them to Racer. As you know, most sim games have 3 LOD's (Level Of Detail), so I just tried to convert the best level (LOD0 in most cases) to Racer.

    The FM3 cars are so precisely designed, that the vertices get merged "automatically" by Racer engine in the "optimization/conversion' process.

    My question is quite clear, is there any chance of seeing a better vertex precision for Racer in future versions ?

    For sure, I could redesign the cars & remodel the curves regenerating all surfaces to a lower polygonal result, but my idea as always in life, is to facilitate my work to increase productivity & efficiency....

    Hope Ruud or Mitch can give me an appropriate answer to that kind of issue.
  2. How close are the vertices that are becoming merged?

    I'm sure I've had verts very very close to each other before without problems. Hmmmm. Any screenies of the offending areas in-sim?

    An test is to export at say 10x larger than it should be, optimise, and get the DOF, then use the scale {} option in the car.ini file to bring it back down. Using that method (for now as a test) should allow you to avoid any DOF optimisation issues with regards to pixel coordinate accuracy, since I doubt the scaling method will do anything except scale the coords down.

    I agree, an easy life would be good when converting.

    I'm wondering if this problem is something else though... it'd be interesting to know the vert precision Racer uses... gonna do some tests now :D

    Can I get a 0.1mm sided cube to appear as a cube :)

  3. Is there actually a problem with the verts being welded though or are you just noticing a large number of them in the popup after the optim.?
    Because F3 uses a lot of different parts for their damage model. i.e. the bumper would be detached from the mudguard so there'd be X verts being welded across there. Probably about 100 or so IIRC. Remember that that'd happen for each panel, totalling up to well over 1000 possibly.
  4. The scaling solution went thru my mind but didn't try it yet, so I'll see if that resolves the problem.
    That kind of issue didn't happen with Shift cars, which are really low-pol modeled compared to FM3 cars.
    Just a precision, a full FM3 car contains about ~ 150K - 200K polys, about 2/3x more pols than Shift cars, for sure at LOD0, all included.

    Before exporting FM3 cars to Racer, I naturally get rid all underneath collision meshes, so by default, the preparation process is quite huge.

    The result once the FM3 cars LOD0 are exported to Racer is simply a "screwed" mesh in areas where vertices are positioned tightly in space (detailled areas).
  5. What tools do you use, btw? And can you perhaps post a few screens of your problem. I have a little hard time understanding the situation, but I would really like to help here :)
  6. Cam, can you send me a mesh over to play with... I'm tempted to try replicate this too.

    I guess the cost of such precision is bigger files., and more memory used to hold them.

    Considering such precision is only important when poly levels are very high anyway, then these meshes are going to use huge amounts of memory... I'd like to know that we actually need that much precision before we starting increasing memory allocation to meshes when, so far at least, I've never seen this problem on any model in Racer... even fairly HQ ones like Cam's WIP F458... Hmmmm

    Will it ever be important to resolve details down to say 0.1mm... maybe in a photomode sense (which FM3 does, and only seems to use these super HQ meshes in), but for in-game driving?

    Still, it will be very interesting to know the limitation of Racers engine in this regard for future reference :D

  7. I did a try scaling up 10x bigger, as you can see, in detailed areas, Racer Modeler screws my prepared mesh. What you see in Maya looks the same in 3DS Max...hmmm

    Quite frustrating...


  8. Have you tried exporting from 3ds max using the DOF Exporter plugin and without using Racer Modeler?
  9. Will it ever be important to resolve details down to say 0.1mm... maybe in a photomode sense (which FM3 does, and only seems to use these super HQ meshes in), but for in-game driving?

    Forza 3 use these lod0 meshes when they are on photomode.
    All the rest only lod1 is used.
  10. Yeah I thought so.

    The photomode stuff looks VERY smooth and high detail in FM3, but the in-game meshes are clearly an LOD down in comparison. You can really tell when you take a 1080p photo live, vs a photomode image!

    It makes you wonder if FM3 has a vertex accuracy issue too, and in photomode it uses a more precise system to load in the elements for the rendering of individual frames?!

    As said, Ruud was concerned making it so any single mesh was too high on the vert count due to memory issues, and vert accuracy would probably be a similar thing too. These meshes would just start to take up WAY to much memory for accuracy we see maybe 0.1% of the time in super-close ups... which imo, deserve only to be in a photomode type thing, not in the realtime stuff.

    That said, this is clearly something we need to know so it can be avoided.

    I think the solution is to use the best realtime mesh, rather than the photomode mesh... if that is the case of where the problem is :D

    It's not like the in-game meshes are lacking at all... imo they are damn good as well :D

  11. Forza 3 Has some limitations too.
    Look at this.
  12. It looks nice in a photo, but that level of detail is absolutely pointless for driving around. It just hogs resources and HDD space imo.

    That said, it does look nice. But these photo mode LOD models really shouldn't be converted for realtime Racer use where you just can't appreciate that detail level.
    FM3 looks great in motion, only in the HD stills do you see the cars are not THAT high detail, but it's ample for realtime use :D

  13. look at that 'automatic welded' part on the side of headlight.
  14. Hmm

    I admit it's a bit abused, but since we have cameras (cars + track) which can zoom really near to the car, you understand why I'm requesting a better vertex precision. From cameras, you can see every defect & nowadays with our CPU/GPU insane power, we could focus more on the rendering issues ingame (like FM3, GT5...) than focusing on resource optimization...imo

    Anyway, LOD1 instead of LOD0 for the body & LOD0 for the interior is one way of getting really nice results (already tested) too, the other LODs (LOD2 - 5) are really too low-pol, still they could be easily integrated in Racer (always wanted to try that feature, will do it soon)...

    I like perfect & good looking things, imagine you play Racer on a surrounding config, you would see any little defect ! Seeing lines instead of curves isn't nice at all....
  15. Hmmm, I'm happy to see people pushing what is possible.

    BUT... just as an example.

    My test track, and my old M3, or Endo's Nissan GTR (better looking than my M3 by a long way, and amply good looking imo for a realtime game), and I get a good 110fps.

    Throw in Cam's WIP F458 (lod similar to a photo mode FM3 model) and the fps are about 55-60fps.

    In my view that is not acceptable when you can't really tell that the car is that much better until you zoom in so it fills the screen and start doing fancy screenshots etc.

    Now, when you do that, that is fine too, but then there is a very stark issue of car quality vs environment quality!

    I'm just worried that we are benchmarking GPU speeds with one car on a very small light track like Carlswood...

    Just imo, from what I have seen, high density meshes seem to have a big cost and a very small visual improvement tradeoff *in game*

    That is the core issue for me right now. In-game I think Endo's GTR is ample. But in menu's and screenshots etc, I do accept that these fancy cars with opening doors and super smooth panel gaps and all that do look impressive. Just I'm not sure Racer should try compete there when Racers strength, by a long way (despite it looking good too), is it's driving and environments.

    Just my 2p.

    I plan on having an LOD0 model for my Z4 at ~ 70-80k polys, but the LOD1 one will be ~ 25k polys and look ok for the most part, but it's nice indeed to have a high poly one for the extras like photomode or menu screens etc... not sure how that will work though with vert accuracy. I'll try do some more tests tonight and see if I can see problems arising :D

    I'm sure Ruud will mention memory allocation though. There was an issue with verts per geom, and if Cam's F458 FPS hitter is anything to go by, moving to 16bit for more verts per geom, and then more vert accuracy too, then these big meshes are going to be even more expensive.

    I wonder then how FM3 etc manage to use these meshes effectively... are they literally used ONLY in photomode/menus, and get nowhere near the realtime engine?

  16. Perfect example also of just being a bit lazy from an authoring point of view.

    The bodywork has the headlight washers in mesh, but to me I could live with a normal map decal there, and then spend more time on that ugly, what looks like flat mapped, grille next to the light.

    All the realism had from modelling that headlight washer cap is lost with that naff grill shader/texture. If that is photomode, THAT would be better with a polygon and normal mapped (for the edges of the polys) solution.

    Imo anyway. Obviously it'd take longer to make, and FM3 is made to a cost, but in Racer we don't have such limitations really. We should make stuff like Stecki did, with love and attention to every polygon and every pixel, making them all count!

  17. The LOD0 meshes are originally NURBS surfaces from a CAD program. So best suited for rendering purposes only.

    I was wrong, Racer DOESN'T merge/weld automatically vertices together after some re-checking.
    I mixed the LOD1 exterior + LOD0 interior + LOD0 wheels & done some proper 'traditional' SG/mesh/UV surgery & here's the first 'satisfying results' from a good perspective, car is about 60K without wheels, pilot/helmet & steering wheel.


    Cheers !