I would need to see the problem to give a good answer as I'm not sure what I'm envisioning in my head matches what is being described.
Knowing how the sphere is modified is important. Unfold tries to evenly distribute the texture space of the texture to match the topology while minimizing stretch/compression. If the poles have elongated triangles, then that will obviously play into the distribution of the unfolded texture (in theory it shouldn't, but in practice it does). One way to mitigate that issue is to add vertices and perpendicular edges at regular intervals along the elongated edges to closely match the spacing of other edges around the rest of the mesh, but that will have limited influence on the result and is more of a brute force technique. Unfold is also a flawed tool as even simple cases come out distorted. for example, get a primitive sphere and imagine it's the Earth. Place a vertical UV seam down one side at the international date line, then two more seams at the arctic/antarctic circles. Deselect the vertical edges connecting the circles to the poles. Now unfold the mesh. Notice the sphere is splayed in butterfly fashion, but one half is larger than the other and slightly off kilter in alignment with the texture editor. The circle at one pole is often (but not always) larger than the other circle too. These are the kinds of issues you'll battle, but on more complex cases they'll be too complex to solve without resorting to cleanup via pushing/pulling points to correct the flawed parts of the unfold. However, let's put all that aside and look at the goal from the beginning and not the current situation which has a roadblock. An equirectangular projection comes in a few flavors, but most are similar to a cubic projection. The main difference is in how the top and bottom sides are projected. Simple analysis of the problem would suggest one could take a cube and use Catmull-Clark subdivision smoothing to round it into a sphere. That would accomplish nicer edge placement which closely match the meridians of the projection to handle (or fabricate) the texture space. A single vertical seam from pole to pole (despite no physical poles) could be used to unfold and splay the sphere to accept/define the projection, but subtle details may need to be tweaked for a perfect match. Alternately, use rendermap applied to a sphere to capture the external world. The rendermap generated image should mimic an equirectangular projection. You may have to open the poles like the Earth without the arctic/antarctic circles, for example, to adjust the field of view for the rendermap process. Invert the sphere's normals so rendermap points outwards into the world instead of inwards towards the sphere's surface....and of course, exclude self or make the material 100% transparent so it doesn't block the rendermap camera from seeing the world. Since rendermap travels texel-to-texel along the geometry, a high resolution image and smooth surface are really important. I'd encourage you to use a NURBS sphere with view dependent smoothing for best results. you set those in the sphere's geometry approximations PPG. Try setting length/distance/angle values to less than 2 degrees and 0.5 units, activate view dependent subdivision smoothing, and make sure the min/max subdivision limits are increased beyond the default 1,3 (well, just the max. shouldn't have to go beyond 6). If you get sawtoothing at the poles or faceting artifacts in the resulting texture, then it means your smoothing parameters are not set correctly. The reason for using NURBS over polygons is the better interpolation of the shading normal between texels. Rendermap is highly dependent on the shading normal orientation to determine what it's camera points at. When pointing a camera into the outside world, even very tiny deviations in normal orientation can produce big errors in the result. NURBS are infinitely smooth whereas polygons are only as smooth as they are subdivided - and even then approximations at best. Matt Date: Sat, 14 May 2016 17:33:32 +0200 From: "Sven Constable" <[email protected]> Subject: RE: equirectangular uv To: <[email protected]> To fix distortion on the poles, XSI has a special mapping feature called 'implicit' (Clusters/?Texture Projection Def), but this is actually a mental ray feature and doesn't deal with UVs at all. So when exporting meshes you cannot use it, I think. I'm not familiar with Unity unfortunatly, maybe there is a similar feature for spherical projections not using UVs but instead a special projection method (perfect spherical) ? Otherwise, since a sphere always has poles/singularities you will get distortions on them. Workaround could get rid of the poles by deleting the inmost polygons on each pole, duplicating the resulting (open) edge loop, and scale it to zero. Resulting in many point on the same spot. Then relaxing them in the texture editor. Results could be ok, not sure. Maybe I'm overcomplicating it. Matt Lind needs to chime in :) Can't you use cubic mapping? That should avoid the problem in the first place. sven ------ Softimage Mailing List. To unsubscribe, send a mail to [email protected] with "unsubscribe" in the subject, and reply to confirm.
