Parallax Occlusion : A possibility?

[lysol]

While this looks great and all, the part that is interesting is really: Will this put less load on the computer compared to just making the models look this detailed from the beginning?

I imagine so, since it requires less memory and automatically lowers detail at a distance by simply not generating as much extra detail, rather than switching between different models.

I mean, to me it feels like it must require a lot of calculations for the computer to add those details from simple height maps, can this technique really be that much more efficient on presenting high detail compared to a "normal" 3d model?

The calculations are handled on the GPU, which happens in parallel. It's trading more processor usage for less memory, but the reduced memory usage and use of parallel processing can work in its favor.

Interesting. Thanks!

[Ravenwing]

While this looks great and all, the part that is interesting is really: Will this put less load on the computer compared to just making the models look this detailed from the beginning?

I imagine so, since it requires less memory and automatically lowers detail at a distance by simply not generating as much extra detail, rather than switching between different models.

Can tessellation work in reverse? As in, can it also decide to lower details below the original mesh? I imagine this would eliminate the need for distant static generation like MGE makes you use, making the distant lands/statics implementation much easier.

[wareya]

>As in, can it also decide to lower details below the original mesh?

This is a "Hard" problem and it's normally done by modelling software that's aware of smoothing groups. Indeed it works somewhat poorly on arbitrary models and modellers aiming to use automatic LOD creation have to design their meshes in such a way that it makes the modelling tool's life easy. Grids of quads are good. Complex triangle patterns are bad. LOD generation requires a lot of procedural decision making which makes it seem like something that would be very slow on GPU.

I think the distant land generator has an LOD generator already, that should give you an idea as to what it would look like.

[GeorgeTheWarp]

Okay, I'm really sorry to bring an old thread back from the deads, but hey, with all that hardware tesselation talk, I got something to add.

I feel like hardware tesselation might indeed be a great thing, and it is becoming standardized (OpenGL 4) and it is supported by ore OpenGL 4 AMD cards. this is a great news for indie game devs because they'll be able to use way simpler shapes, and permit distant LOD without having actual LOD, to still look good. A slider for the tesselation could indeed be used, and models that are simple, the closer you'd get, the more "real" they'd look.

So using a 5 edge cylinder to make a barrel would now be possible, for so long as your normalmap and heightmap are good enought, and a lot of time spent on making the model could now be spent on making better textures and bumpmaps. No more would the 3D artist have to make high poly meshes, perphaps even for character. (but you'll see me advocate for cartoon character meshes anyways, because they stand the test of time)

And not only that, but 3D grass wouldn't be needed, so that's a great news, both for performance and hardware availability.

Either ways, I don't see this being supported very soon, but it'll hopefully be someday.

[AnyOldName3]

Generally, in order to generate a really good heightmap, you need to have a very high poly mesh and a lower poly version you'll be using in the actual game. You use a tool to bake a heightmap based on the differences between the two models, and then as you add more parallax occlusion, tessellation or whatever, it ends up more like the high-poly mesh. If you don't do this, then everything tends to just end up being really round.