7

I have implemented global illumination using the Monte Carlo method, using the scratch a pixel tutorial as a guide. My final image renders very noisy! The example below is at 64 samples, I have previously used as high as 512 and its still very noisy.

Any ideas what the problem could be?

64 Samples, 1 Bounce

Edit: Here is the output with 128 samples and 16x Super sampling,resulting in 2048 samples. Still lots of noise!

enter image description here

Arthur
  • 347
  • 1
  • 4
  • 14
  • That's just how it is. Mine works much the same. – Alnitak Apr 17 '17 at 10:08
  • Are you sure it supposed to be this bad? I have tried to combat the noise by using 16x supersampling (grid algorithm) and 128 samples but its still bad! – Arthur Apr 17 '17 at 10:49
  • Mine needs a few thousand samples to get acceptable noise levels. FWIW, I don't use a grid for supersampling - I just pick floating point x and y samples and then just add those into the right pixel bucket for those ray coordinates. – Alnitak Apr 17 '17 at 12:58
  • p.s. http://raybellis.github.io/RRT/ – Alnitak Apr 17 '17 at 13:10
  • Just uploaded a new image with 2048 samples, still lots of noise on the floor. I know its supposed to be noisy but surely not this much? Could it be to do with the fact that rays that exit the box contribute a sample of 0? – Arthur Apr 17 '17 at 13:47
  • 1
    The rays exiting the box certainly won't help - putting a large sphere around the world can help a lot, although it will of course also affect the overall illumination inside the box. – Alnitak Apr 17 '17 at 16:00

1 Answers1

9

Path tracing is pretty noisy; it's the nature of the algorithm. Consider this example from Wikipedia:

enter image description here

The top left image is at 1 sample per pixel, and from there (left to right, top to bottom), each following square doubles that. So the bottom right square is at 32768 spp.

There are other, related algorithms that can reduce noise for the same amount of computation:

Daniel A. Thompson
  • 1,904
  • 1
  • 17
  • 26