QUOTE(Philip Grah @ 05/31/07, 02:49 PM) [snapback]267156[/snapback]
You should just map onto the normal vectors, not reflection vectors. The reflection sphere map is actually the front half of a sphere, and it represents the reflection seen across the normal vector of the sphere. Mapping the reflection sphere onto reflection vectors is redundant, so you get the full 360 degree reflection mapped twice onto the front 180.
Thanks for replying!
If the shot is static then it may be fine to use normals. If the object or the camera is moving, the reflection vector is needed. As you are saying, using the sphere is redundant due to the mirroring, which is why I would like to find some way to use a lat/long map.
Edit: I want to re-write my response.
If you map the reflection sphere(ie. mirrorball image) onto the reflection vector pass, you get the full 360 degree image mapped twice, as you say. If you convert the mirrorball to a sphere map (180 degrees) you get the right image warp but it will mirror around the seam which you will see if the camera moves around it.
The same applies if you are using a world normals pass and want to do an environment lookup to mimic IBL, assuming you supply a 180 degree sphere map. It will look right on the front 180 but it will mirror on the back because you have thrown away the other 180 degrees in the image to get the correct warp.
If you map the mirrorball as is (ie. not converting to a sphere map) using the world normals as you suggest, you can see that this is wrong with things like a ground plane because it only looks at the up (+y) vector and returns a solid colour instead of a mirror reflection.
