LODs - what is the best approach ?

Discussion in 'Car Models | Physics' started by luchian, Oct 29, 2017.

  1. luchian

    luchian Administrator Staff Member

    Joined:
    Jun 3, 2014
    Messages:
    3,007
    Likes Received:
    1,493
    Somewhat influenced by @FuzzyDice 's question, it got me thinking: how to tackle LOD creation ?

    Generally speaking, LODs are used in games to help with the performance. Meaning that once an object is far away from the camera, there is no point in using a high-poly version of it, since user will never see it. And it would only force an unnecessary amount of calculations that would only slow the game down. Therefore, simpler versions of objects make more sense, and linking their appearance to the distance from the camera makes perfect sense.

    Enter LODs. Which btw, stand for Level Of Detail. And btw, btw stands for by the way :p

    So, using LODs means you will have to create multiple versions of the same object. Which is the best way to do that ?

    There are mainly 2 ways on going about it:

    1/ Creating a model in steps, keeping in mind the polygon "target" at each step. For example, doing the best you can:
    First in 2-3k tris for LOD D, then save.
    Then 10-12k tris for LOD C, then save.
    Then 20-25k tris, save.
    Then 125 k tris.
    The end.
    PROS:
    + you would have full control on what yhe final mesh looks like
    CONS:
    - it would take forever
    - requires perfect planing from the begining
    - you would most likely need to do UV mapping multiple times​

    2/ Create the best model that you can, without considering LODs.

    2A: Once done, do the UV mapping. Then use decimate-like modifiers, to lower the poly count to target values.
    PROS:
    + fastest method
    + not only fast, but also automatic
    CONS:
    - it can require some tweaking with parameters
    - it may never look quite the way you hope​

    2B: Once main lod done, remove loops/vertices by hand.
    PROS:
    + better results than with automatic methods
    + full control on what is important and what not
    CONS:
    - it takes longer​

    Personally, I go for the "automated" method on the first try. And only correct where visible problems. But I am sure that's just the simple way, not the best.

    Useful links:
    Simplygon (if using max, maya, etc).
    Maya automated tool

    Related (sort of):
    Instant field-aligned meshes

    This is merely a start for discussions, feel free to add if you know about other tricks or best practices.
     
    FuzzyDice, Pixelchaser and Willy Wale like this.
  2. LeSunTzu

    LeSunTzu New Member

    Joined:
    Oct 16, 2017
    Messages:
    7
    Likes Received:
    6
    If you are boxmodeling there is an advantage in doing the uvmap before cutting out car parts: uvs will be aligned and more friendly to the many guys who skin only in 2D.

    Keeping uvs for all LODs is a must for objects that must keep the same texture like carpaint skins, but for others it can more efficient to have another texture. For example a building with an outside circular staircase: the only way to simplify the railings and stringers is to replace them by a cylinder with a transparent texture.
     
    MLT and luchian like this.
  3. AccAkut

    AccAkut Active Member

    Joined:
    Sep 4, 2017
    Messages:
    121
    Likes Received:
    104
    What I do is a mix of both 2s. What you did not write there is just straight up deleting whole submeshes. Like with my KRB Audi, LOD B loses all the underbody stuff (engine, drivetrain, suspension etc). Interior is limited to above window line, all LOD A transparent wire gauze parts lose their transparency, wheels lose bolts and inner lip, front and rear lights lose bulbs and NM layers. LOD C goes further with this, also I use a new simple matt black material on most objects to cut down on DIP.

    LOD A is 317 objects, ~300k tris, LOD B 114 objetcs, ~85k tris , LOD C 67 objects, ~40k tris.

    Might add that my LODs are nowhere near those guidelines, but having them at all seems not to be a standard practice on mod cars.

    With my current project (a fantasy exeskeleton car) I plan on pulling the Bezier curve skeleton up from a backup file, and rerender it at minimal resolution for LOD B and C, as its probably heaps easier than decimating the 50k high res cage back down.
     
    luchian likes this.
  4. garyjpaterson

    garyjpaterson Member

    Joined:
    Jun 28, 2017
    Messages:
    97
    Likes Received:
    99
    Not a lot of experience with LODs myself, though everything I've done so far has been 2b - all done by hand.

    To be honest its not too bad, the only thing I hate about it is the fact by that stage in the model I'm not using symmetry modifiers, so you have to remember to make the same changes at each side.

    I think next time I attempt LODs I'll try some automated processes and see how it goes - thankfully for these Ferrari's I'm not doing the LODs so I don't have to worry about it :D
     
    luchian likes this.
: lod, lods
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice