Water Reflection and Refraction with normal map with out pixel shaders

quote:

---

Original post by Yann L
And perspectively projecting it back onto the water surface, while distorting the texcoords. Here again, you can either do it per-vertex (the distortion will only be influenced by the vertex normals) or per-pixel (the distortion will be influence by a bump-map, needs pixelshaders).

---

quote:

---

Yann, mine flicker because most of my plants use alpha testing. So, they have this tendency to flicker anyway, but they don't flicker that bad. However, when they flicker in a 256 texture, that flickering is greatly amplified, especially in high resolutions...

---

quote:

---

Why can't I get similar results? But then again, I'd have to see it animated....

---

quote:

---

How to distorting the texcoords,per-vertex?
Use the View Space vertex normal.x and normal.y?

---

    
// N is the vertex normal, TC the 3D projective texcoords (ie. the vertex position, that gets projected to screenspace texcoords)


// get the displacement offset

float dofs = VIRTUAL_PLANE_HEIGHT * N->z;

// Displace the texcoords (vertex position) by the offset, in the direction of the normal

TC->x += dofs*N->x;
TC->y += dofs*N->y;
TC->z += dofs*N->z;

// The projective texcoords are now floating above the surface, on the virtual plane.

// We don't want that. Correct the height. Little hackish, but works.

TC->z -= VIRTUAL_PLANE_HEIGHT;

// Now you can feed the texcoords TC to the projective matrix (in the texture matrix), and render the surface with the projective texture.

// ...

    

quote:

---

Original post by Yann L
Doesn''t really matter, you can do it in world or in eye space. Just make sure to stay consistent in your coordinate frames, ie. if you do the distortion computations in eye space, make sure to displace the eye space texcoords.

I would suggest doing the distortion in world space, and doing the world->eye->screenspace transforms and projections with a single texture matrix, in a vertex shader.

---

quote:

---

Original post by Yann L
OK, let''s start from the beginning.

You have your reflection map in unit 0. Now, you need projective texcoords to adress unit0, so that the texture reprojects the same way into screencoords that it had been ''captured'' before.

The reflection texture was rendered from the camera viewpoint, through the camera transform and projection matrix. To reproject the texture the same way onto the screen (and the water mesh), you need to perform the exact same computations again, only with the texture coordinates.

So, for each water mesh vertex you render, loads it''s 3D position also into a 3D texture coordinate set:

TC.u = P.x, TC.v = P.y, TC.w = P.z (Eg.1)

Now, the texcoords equal the 3D worldspace position of the water mesh vertex. Now you apply the same camera view and camera projection matrices to those texcoords, as you previously did to render the reflected geometry (except for the reflection matrix, of course). The easiest way to accomplish that, is to load the combined camera/projection matrix into the texture matrix.

If you do that, without any further processing, you will get a perfectly calm reflective water surface. Now, consider Eq. 1 above: if you displace the texture coordinates TC (in worldspace here !) with the worldspace normal, before transforming them by the camera view and projection matrix, then you will get correct behaviour.

Example: in worldspace, a normal of (0,0,1) (assuming z pointing up) denotes a calm surface. Let''s also assume, that the virtual plane is at 5 units above the surface:

dofs = 5 * N->z = 5

TC->x += dofs*0 == TC->x += 0
TC->y += dofs*0 == TC->y += 0
TC->z += dofs*1 == TC->z += 5

TC->z -= 5

As you can see, no changes are done to the coordinates. That is correct behaviour for calm surfaces. Note that, depending on your worldspace coordinate frame orientation, you might have to flip the sign of the virtual surface distance.


BTW, I''m planning to give an IRC lecture about realistic realtime water rendering shortly (exact date/time confirmation pending). You might want to drop in.

---