I'm trying to implement an ocean scene with C++ and DirectX11. Currently I have a projected grid, Gerstner waves and a basic shading. My problem is that when I aim my camera horizontally, so I can see the water horizon, in the distance, the projected grid becomes insufficient, even at high vertex numbers. These screenshots illustrate the problem:

shaded water surface

wireframe water surface

I know the cause of the problem is in the concept of the projected grid (the grid is detailed near the camera, rough far from it), but there has to be a best practice to solve this.

Any ideas?


I believe a common solution is to split the camera transform used to project the grid from the camera transform that is used to render the grid. At perspectives close to top-down, the two cameras coincide, but as the viewing camera gets close to a horizontal perspective, the projection camera deviates and tries to keep a minimum inclination, i.e. it hovers somewhere above the view camera and looks down slightly.

The tricky bit is making sure that the field of view of the projection camera always covers the region of the scene seen from the render camera. I don't have a resource at hand that details how to compute the appropriate transforms, and it might be tedious to derive by hand.

A different solution is to grab the signal processing toolbox: The artifacts seen in your image are essentially aliasing, caused by insufficient sampling of the wave heightfield by the projected grid. Therefore, one solution is to filter the heightfield appropriately, depending on the projected area of a grid cell. I believe this is used in offline rendering of oceans, and it essentially makes sure that waves at the horizon go flat. However, I'm not sure how feasible this is in real-time rendering, since you would need high-quality anisotropic filtering to make this approach look reasonable.

  • $\begingroup$ Thanks for the advice, I chose the lazy solution for now. I use a function in the vertex shader that determines the wave attenuation from the distance from tha camera. $\endgroup$ – Gábor Szalóki Nov 5 '15 at 18:52

You can be both realistical and real-time. the secret is to change representation each time the information get under the Shannon-Nyquist (i.e. grid) scale: from geometry to normal maps to shading models. This paper is made for you: http://maverick.inria.fr/Publications/2010/BNH10/index.php (see also Yoube videos)


Some software like Maya, solve this by using a polar (or actually cartesian that turns polar at a distance) much in the same way as you grid centered on the camera position. This setup adds more detail where it counts most Then they rely on the shaders normal processing at further ranges. There is room for improvemenet offcourse. You cold modify this approach a bit, and have any other shape that increases the mesh density towards the camera. The benefit of this is you can stretch the effect up to the horizon without worrying about the seam.

The trick to not get the displacenent jumbled up in this case is that you gradually reduce the displacement as you move further away. You then just use normal modification in the pixel shader as you get further. This is easier to filter than having to filter an accurate shiluette edge. Also if you can see that far away then your vawes are likely sufficiently flat anyway.


The technic what Benedikt mentioned is explained in Section 2.4.1 of this thesis.


Implementing this should solve your problem.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.