3 replies to this topic

[Polygon]

Hello again all!

I have some misc. questions about indoor / outdoor rendering. I was hoping someone could clarify or explain some things to me.

Ourdoor Rendering:

1. When rendering outdoor scenes which is the best option: Heightmaps or Geometry made in a level/model creation application ( e.g 3dsmax ).

2. To render outdoor scenes, how does one get those smooth transitions between surfaces. For example when grass ends and sands begins.

3. Is it better to utilise static geometry as created with question one or should one use an algorithm such a ROAM for tesselation.

4. How can one manage large terrains? Is it wise to split them into groups / partition them.

5. How can one perform occlusion culling with large terrains?

Indoor Rendering:

1. How would one manage the transition from an indoor to an outdoor environment if one were using heightmaps?

Thanks!

[anubis]

1.) there are some good algorithms to generate height map data for you. i doubt that anyone can model realistic rock formations by hand (or would want to).

2.) You could use multitextureing to blend together a grass patch over a few rocks and have different grass textures for different heights (transferable to sand, rock transitions as well).

3.) Depends on how much data you want to display... considering 4.) i wouldn't use static geometry.

4.) level of detail algorithms can help

5.) if you draw a slope on a piece of paper it's easy to see that anything behind the slope will be invisible. therefore you can place "anti-portals" where big mountains are (by hand or automated) and cull everything that is visible through these portals (if you aren't familiar with portals, just imagine that you place a shape where the mountain is and think of that as an anti-window where you can see everything except what you see through the window).

[Rofar]

Not sure I can add much different than anubis' reply but..

1. It really doesn't matter how you create your heighmap data, in the end you will just be loading vertices to be rendered.

2. blending textures (like anubis said)

3. you will want to use some kind of lod technique but you can still start with static geometry

4. split a single large terrain into sections, only render sections within the view frustum.

5. occlusion won't buy you much with terrain rendering if you do number 4 above. I'm not saying don't consider it, I'm just not sure how gain you can get for the effort to do it.

Indoor rendering
1. Not really sure. I haven't done this yet in my engine.

You can get complete answers to most of these questions by grabbing a copy of
'Real-Time 3D Terrain Engines using C++ and DirectX' by Greg Snook and published by Charles River Media.

[Michael]

anubis said:

1.) there are some good algorithms to generate height map data for you. i doubt that anyone can model realistic rock formations by hand (or would want to).

You are wrong in the second part. Also, consider the limitations of height maps.

Quote

3.) Depends on how much data you want to display... considering 4.) i wouldn't use static geometry.

Considering 4., I would use static geometry. Note how there's a continuous shift from all-CPU algorithms to all-GPU algorithms (ex. ROAM->Geometry Clipmaps) which makes a lot of sense).