How does the Arnold spherical camera work?

[zachcoley123]

I’m wondering how the Arnold spherical camera renders scenes? Is the the FOV or is the pixel grid that each ray is traced through being wrapped into a sphere somehow? I don’t think it works like real 360 cameras by using multiple cameras aiming in different directions? Can someone shed some light on this? I’d like to build my own version of it using either OSL or with a Arnold custom .dll plugin so I can experiment with it a bit more. Thanks.

[zenop · Accepted solution]

You're in luck, there's documentation for this! Note camera's can't be implemented through OSL in the arnold implementation.

https://docs.arnoldrenderer.com/display/A5AFMUG/Creating+a+Basic+Spherical+Camera

[zachcoley123]

Thank you! I’m also know looking into building my own cubemap projection style camera as after some more testing I found while good the spherical cameras have more distortion at the top and bottom. Which the vertices of a polygon sphere would all join as I’m rendering it with a spherical camera and then reapplying that rendered image to a 3D sphere. These issues are due to the 3D spheres UV’s and the way the camera itself captures the scene. But cube map projections would remove this, even though it would require rendering from 6 cameras, one for each face.