I must say I like the idea behind the project.
You mentionned earlier that you can for example, modify the details of objects such as trees, depending on the distance with the player. I wonder how you can "switch" from "low details" to "high details", without having the player realize the change. I mean, let's say you use a billboard to render a tree far away, then, if you're near it, you expect it to be detailled, so how do you take care of the transition ?

I started out using billboards for this.  As each 3D object was first created, I rendered to a special render target and took snapshots of it from several angles.  These bitmaps were then used as billboards when the object was far enough away.

Ultimately, I didn't like the effect.  It was complicated to implement, the angles were never quite right, and it was difficult to get the lighting right under all circumstances.  And, of course, transitions sucked.  I was about to implement a fade-between approach, but was already unhappy with the complexity of the code as it stood.

But then, I (finally) realized that I could get the computer to do the work.

So, rather than use billboards, I had the system generate a low poly version of every object.  With the trees, this was just a matter of specifying the number of sides and segments of each branch (all objects in the game are generated procedurally).  At distance I use a low poly version (3 sides, 2 segments) and up close, I use the high poly version (6 sides, 4 segments).

-db

Hmm.. good suggestion on optimizing the the collision detection Terra-Infinitum. I had used a simple flag to mark moviing/static objects and another flag to mark if the status of the sprite has changed. This would allow me to skip the collision test for non-moving objects but still requires to test all moving things against static objects and other sprites. I think it is still not good enough.

Btw any hints on how to rank the object so that the closest ones get tested for collision? I think once they are ranked I can limit the collision test to a limited number instead of complete list (e.g. 20 tests assuming that a chracter could practically have not more that 20 other object close to it).

On the draw optimization another thought I had was that since the mesh culling etc. are perfromed at GPU and it might be much faster than if we filter the objects at game engine (i.e. filter it at CPU cost). But again if we dont filter it then mesh draw loop will waste CPU cycles. Maybe it is some thing that needs to be tested practically to get a definite answer.

Xia:

Btw any hints on how to rank the object so that the closest ones get tested for collision? I think once they are ranked I can limit the collision test to a limited number instead of complete list (e.g. 20 tests assuming that a chracter could practically have not more that 20 other object close to it).

Hi, well, you can just seperate the world into small areas. Then, you use bounding boxes to test whether the player's inside the area, if so, then test with all the objects within the area.

@Daaark:

Yes, AutoCAD was originally created for architectural design, but so was 3ds Max.  It wasn't always owned by Autodesk.  It was created because AutoCAD didn't provide that high of a standard for architectural visualization and programs such as Revit hadn't been created yet.  It wasn't until a few years ago that 3ds Max was bought out by Autodesk and became popular for game design.

As for Maya, it's not that it uses polygons or NURBS that makes it unique, it's the "Subdivision Surfaces" (SubDivs).  That is the great feature in Maya, a combination of polygons and NURBS, that allows the modeller to create smooth edges, using these equations built into the program.

And there is only so much you can do with the renderer.  It is up to the modeller to make certain decisions, as it is the model that contains the data that represents it.

Antoine:

Xia:

Btw any hints on how to rank the object so that the closest ones get tested for collision? I think once they are ranked I can limit the collision test to a limited number instead of complete list (e.g. 20 tests assuming that a chracter could practically have not more that 20 other object close to it).

Hi, well, you can just seperate the world into small areas. Then, you use bounding boxes to test whether the player's inside the area, if so, then test with all the objects within the area.

Antoine:
Ofcourse.. why didnt I think of that.  I love simple solutions.  Thanks.

Hi.

I'd started typing up a post about LOD and large maps, and then realised I'd covered most of it already in my QuadTerrain thread! See my signature for the link.

LOD is vital in a large-scale map: a game like KOTOR, with small 'zones,' can get away with the brute force 'render-everything' approach, but the larger scale your map is, the more LOD you will need. The ultimate example (and one which get's used far too often) Crysis, LOD's everything: terrain, textures, vegetation, objects, physics, shadows... even AI gets LOD'd. When you start looking for it, it immediately becomes apparent.

Also: I'm a fan of 3ds max. I started with max 5, back when it was still owned by Discreet. :D