Hi there

I was wondering how the LOD for really big objects/ships (planet or moon sized) would work.
I was picturing a huge ship that looks similar to an aircraft-carrier .. except it is as big as a moon, with lots of generated detail.
Then I thought if you used traditional LOD mechanisms the whole thing would have to be loaded into memory when you get close.
If it has a lot of detail then it could be a million polys or more at that scale, especially if the object is procedurally generated.

So I went looking for ideas of how one could deal with this scenario and found
"View-dependent refinement of progressive meshes" http://research.microsoft.com/en-us/um/ ... roj/vdrpm/

So the question I have is, do you guys think this could be realistically done with a UV mapped object and in real time?

Josh is using low poly impostors to achieve this for distant LOD objects.

N810 wrote:Josh is using low poly impostors to achieve this for distant LOD objects.

Josh's method only works per-object.

Either the whole object is lod'd or it isnt.

The article is about applying different LOD levels to different areas of the same object.

Which would be very useful for really big, complex objects, like big ships and stations.

(Planets not so much, a perfect sphere doesnt need much LOD )

RaXaR wrote: So the question I have is, do you guys think this could be realistically done with a UV mapped object and in real time?

Well, the paper is from 1997/8 and they claim that it only used 15% of their computers processing power.

Thinking that the touch screen controller on my smartphone likely has more FLOPS than that workstation they used i guess that it should be very possible today

I honestly don't know of LT will have anything that will require progressive LOD.

Most games simply use 2 or 3 levels of LOD, swapping them out at specific distances.
LT could use this same method for basically everything.

When you are at huge distances, a simple 3 sprite impostor can be used.
At closer distances we can use the model with very low texture resolution, and lower detail. Seeing as the models are being generated procedurally, I would assume that it generates the major detail parts, then the minor, then the tiny. So simply stopping it before it generates the smallest detail parts would be sufficient to create LOD models.

But progressive LOD is more useful for objects that are immense ad are at long distances and close up at the same time.
Like a Star Destroyer when you are standing at the engines, the mid part would be low LOD, the tip lowest LOD.

From the sounds of it, LT will be fast enough to not need to bother with this.

Silverware wrote: From the sounds of it, LT will be fast enough to not need to bother with this.

LT is so fast because of such tricks

Trench runs: much better with moon-sized hardware.

If this tech can make that sort of epics more awesome and viable, then it's on my wishlist.