Open astrofrog opened 10 years ago
Alternatively, do a first pass to find out the volume of cells that contribute to the image, to have an idea of the rough sampling to use, then top up cells to have at least n photons, then use that to do the raytracing as we currently do.
The latter depends on viewing angle of course, so this should be repeated for all viewing angles?
Raytracing is currently very inefficient when there is a large dynamic range of cells because cells that appear larger in the images require more photons to be less noisy.
One option is to raytrace from the image through the grid and keep track of how many times each cell contributes to the image, then to go and sample from cells that are missing from the image? This could end up being extremely efficient.