Hi Kim, I used the same 15x15 grid tiles and 64x64 using a pagedLOD system. The FFT height maps were pre-generated. My biggest problems were seams in the tiles and trying to introduce reflections. I handled the seam problems between tiles by slightly overlapping them at the edges, not an ideal solution but it served our purposes based on the amount of effort we could put into this. I also used a distance based system so only the fragment shader was active until you got to about 100- 200 metres away and then the vertex shader would kick in and give form to the waves. Got around 25 fps on a NVidia 7800, 1.8Mhz cpu
I don't know how we are going to coordinate this effort but I can put everything I've done together and post it. On Thu, 2008-11-06 at 17:01 +0000, Kim C Bale wrote: > Right blimey, I didn't expect so much interest. > > In answer to J-S and Paul the actual ocean surface rendering is something I > find particularly interesting. There are many different approaches for this > (Gerstner, perlin noise, FFT etc), each of which have their own advantages > and disadvantages. > > An idea I had a while back at the start of this project was to create a > nodekit designed specifically for ocean surface simulation and rendering. > Something similar to osgShadow, where various techniques are implemented and > can be chosen based on the application. So yes it would be interesting to try > and tie in Paul's work. This type of nodekit I think would be fairly easy to > design. > > The ocean technique I have implemented is based the FFT technique described > by Tessendorf in his paper "Simulating Ocean Waves". It is the same algorithm > that was used in the film Titanic, albeit without the 5.1 billion polygons > they used. > > The technique produces a tileable height map that can be stitched together to > form a large expanse of ocean water. Whilst it provides very nice looking > waves, the fact that it's tileable makes it rather difficult to represent > local interactions with the waves themselves. Unfortunately wave interactions > were not a requirement, so I have not taken these into account. However, > Tessendorf details an approach called IWave that can be used to provide > these. > > These tiles can either be pre-computed, or computed in real-time. One > disadvantage of this technique is the use of Fast Fourier Transforms to > generate the height maps. FFTs are pretty heavy on the CPU and can be a major > bottleneck. However, for grid sizes of 128x128 it is still perfectly feasible > to generate them in real-time. The advantage of this is that you can change > ocean parameters wind/wave direction, gravity, depth etc on the fly. > Personally, I decided to pre-compute as these real-time changes were not > required. > > In my simulation there is no coast line, you are presented with an infinite > plane of sea. So, I have not considered sea/coast interaction but again this > can be achieved with the IWave technique. > > The sea surface shader is quite lengthy, largely because it needs to take > into account both sides of the surface. It uses cube mapping for global > reflections and a projective texture map for local reflections. I use a > series of pre-computed normal maps based on the height maps to add additional > detail and the colour itself is computed based on the angle at which you look > at the ocean surface. Finally a refraction shader is used if looking upward > from under the surface. The whitecaps are a bit of a cheat really, there are > a couple of clever ways of computing these based on the heightmap, but I > found that simply overlaying a foam texture where the wave peaked above a > certain height provided the best results in my case. > > So, regarding performance, I can happily run a grid of 15x15 ocean tiles > comprised of 64x64 height values. As the poly count gets pretty high, I use > GeoMipMapping to reduce the detail of tiles in the distance. Using that with > the pre-computed height maps I can get 55-60 fps which drops to about > 45-50fps when I render the scene twice on a 8800 GTS. There are a number of > optimisations that I haven't made though. The problem with GeoMipMapping of > course is that detail is relative to the position of the camera. So if you > have multiple cameras you have to re-compute the geomipmapping for each one. > I was fortunate as my secondary camera is very close to my primary one so > that was not an issue. > > Aside from the ocean I have: > > God-rays - Created by rendering geometrically modelled rays to an FBO and > then blending them into the scene as a post render. > > 2D fogging equation - a pre-computed fogging texture is used to simulate the > underwater fogging based on distance from surface and viewer. > > Depth of field - Not exclusive to underwater scenes obviously, but does a > nice job of creating a watery blur in the distance > > Wobbly screen distortion - Post render effect, similar to the effect seen in > BioShock, that warps the view in a watery fashion. It's not a particularly > realistic effect but provides a nice underwater visual cue. > > Bubbles - implemented using osgParticle > > Silt - a volume of tiny particles drifting in the water that tracks with the > camera position again osgParticle > > Seabed Silt - the clouds of sediment created when you kick up the seabed. > > Caustics - projective texturing of a series of animated caustic textures > > Waving plants - simple vertex shader that applies a sine wave to underwater > plants to simulate currents. > > I think that's about it. With regard to turning this into a nodekit, > unfortunately I can't do as Robert suggests and release this code as is. The > simulator itself will not be open source, so I'll have to pull out all the > relevant bits and present them for review. > > So in the meantime I shall start extracting all the juicy bits. Once that is > done I'll get back in touch here and discuss with those that are interesting > in helping. > > Thanks for the link to the Ogre library I shall have a look at the approach > they've taken, seems like a good place to start. > > > Phew.. long email. > > > Regards, > > > Kim. > > ***************************************************************************************** > To view the terms under which this email is distributed, please go to > http://www.hull.ac.uk/legal/email_disclaimer.html > ***************************************************************************************** > _______________________________________________ osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org IMPORTANT: This email remains the property of the Australian Defence Organisation and is subject to the jurisdiction of section 70 of the CRIMES ACT 1914. If you have received this email in error, you are requested to contact the sender and delete the email. _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
