David, I've tried implementing points 1,3,4 from your mail. 1 working out fine but hit the performance alot, especially with complicated terrains since it required so much extra geometry. I tried 3 and 4 in a different context when attempting to layer skins on top of a player model to create "clothing". This appeared to pretty quick on my GF4 but neither API seemed to work how I expected. I ended up getting incorrect results in some cases.
Another option may be to generate a dynamic texture once as the quad comes into the "bubble" by combining the textures by hand in a buffered image and then using this texture. This would mean you'd get a single hit as the square comes into view. You may even be able to implement some sort of caching on these textures. Just an idea. Kev David Yazel wrote: >Hello all! > >I am beta testing an unnamed online game. This game is "state of the art" >when it comes to the graphics and I have had a great deal of fun trying to >figure out how they are doing a few things. One of their outstanding >features is the quality of their terrain and one feature in particular >stands out. > >As you move through the world there is a bubble around you which textures >the terrain at amazingly high detail. In most cases this is a detail >texture over a base texture, but in some cases they are doing something >else. In particular their roads have exceedingly fine detail, far beyond >most of the ground. One thing I observed is that the road textures seem to >be built so that there is no "flow" to speak of and would look good flowing >in any direction. Bricks would of course violate this rule, so I am seeing >irregular flagstones for example. Why is this important? Because the >technique I am thinking of would require that the "road" texture be tilable, >repeatable and non-directional. > >1. How I don't think it is working > >I do *not* beleive they are building specially built geometry on top of the >terrain. Because I see this technique applied to other terrainfeatures >needing extra detail I do not think they are building sliced-coplanar >greometry to hug the ground. I think they are re-rendering the same terrain >patch multiple times to produce a final merged texture. > >2. How I think we could do it > >Justin's e-mail on alpha masking got me to thinking. What if, for each pass >on the terrain patch, they combined and alpha mask with an underlying >texture? So lets say that each terrain patch 128m x 128m with a 128x128 >precomputed terrain texture + detail texture (thats what Magicosm uses), Now >lets say that when you are computing the base texture and you were >calculating the road pixels, you also created an alpha mask matching the >portion of the patch where the road was. You then rendered a second pass >for that patch which combined the road texture, say 256x256 (repeated 4 >times across the patch) with the alpha mask. I think you could conceivably >draw just the road against the terrain with a high level of detail. > >3. Multi-pass with multi-texturing. I don't know if this would work, but my >understanding is that if you use more texture units than supported on the >card then Java3d will fall back to multi-pass. if this is true I think you >could set up 4 texture units: base texture, detail noise, high density path, >alpha mask and it would merge all 4 when it rendered. > >tex unit 0=Base texture (modulate) >tex unit 1=noise texture (add signed) >tex unit 2=road texture (modulate) >tex unit 3=alpha mask > texAttr.setCombineRgbSource(2 TextureAttributes.COMBINE_OBJECT_COLOR); > texAttr.setCombineAlphaSource(3, >TextureAttributes.COMBINE_TEXTURE_COLOR); > >4. Using decal group. I am not sure if this would work, but you could have >decal which used the alpha masking technique to only render over the patch, >but only overwriting and updating the non-masked part. So > >First layer: >tex unit 0=Base texture (modulate) >tex unit 1=noise texture (add signed) > >Second layer (decal) >tex unit 0=road texture (modulate) >tex unit 1=alpha mask > texAttr.setCombineRgbSource(0, TextureAttributes.COMBINE_OBJECT_COLOR); > texAttr.setCombineAlphaSource(1, >TextureAttributes.COMBINE_TEXTURE_COLOR); > > >Would either or both work? If we used the multi-pass technique how would >the combiner's be set up? > >5. Fading the "bubble" > >Since this high detail enchancment would only be applied to the level 0 >patches surrounding the user, we need a way to to move the bubble of this >high detail as you move through the world. These patches would be rendering >far more slowly because of the extra passes, so you would not want to just >render every level 0 patch with this technique. In addition you would not >want the road "popping" in front of you as you moved, but rather to unroll >smoothly. Perhaps we could have the alpha mask texture as by-ref and modify >it at runtime each frame by masking out the edges? > >Any thoughts or suggestions would be appreciated! > >David Yazel > >=========================================================================== >To unsubscribe, send email to [EMAIL PROTECTED] and include in the body >of the message "signoff JAVA3D-INTEREST". For general help, send email to >[EMAIL PROTECTED] and include in the body of the message "help". > > =========================================================================== To unsubscribe, send email to [EMAIL PROTECTED] and include in the body of the message "signoff JAVA3D-INTEREST". For general help, send email to [EMAIL PROTECTED] and include in the body of the message "help".
