Does anyone know how to force a large amount of supersampling in Blender or POV-Ray, say 100 samples per pixel, at least? This is for a research project that is heavily dependent on sub-pixel artifacts. Rendering a large image and shrinking it will not work because a few critical pixels will be very bright, and the output values will be clipped. They would need to be averaged before clipping.
Blender seems to support only 5, 8, 11, or 16 samples, since it uses predefined sampling patterns.
POV-Ray does not support non-adaptive supersampling methods, which is necessary because there is a large variation within pixels that may not be apparent from variation between pixels.
Check out the mesh camera in the latest POV-Ray beta (3.7.beta.39). This allows you to specify an arbitrary number of rays per pixel by defining an appropriate mesh (docs are here).
There are some macros and sample scenes included with the dist. If you’re not sure on how to create a mesh camera that does what you want, try asking in the povray.beta-test group at news.povray.org. A simple approach would be to create a mesh representing a standard camera (one face per pixel) and instantiate it into the scene 100 times (assuming 100 samples per pixel), jittering it slightly each time (this can be easily done in a macro).
Also, Jaime Vives Piqueres has a good (but work-in-progress) demo page here. See in particular the section titled ‘Custom Cameras’.