appleseed Users Forum

About SPPM

It’s a very promising algorithm,beacause it generate approximate results very quickly, however there’s some issue with its convergence speed.As far as i know, the photons should not be visible to camera unless it’s caustics.Through a process of ray tracing which is called “final gathering”, the noise of photons could be greatly reduced. This method were adopted in mentalray(which is dead now).


As we can see the noise near the camera is generated by photons, while the fireflies near the lights come from ray tracing. Combining two methods giving a renderer stronger applicability and generality.
Here’s another picture rendered with blender cycles which shows its weakness.
I encourage you guys keep on developing .Overall you did great jobs ! Since cycles mainly focus on flexibility, its lack in speed makes it not suitable for complex scene . Thanks for your selfless efforts! :smile:

3 Likes

Well ,I also noticed that those photons might evolve to be smaller than a pixel, which contribute nothing to the image. If it’true, I think that’s the main reason why it doesn’t converge at all after first ten minutes. When the radius of photon is smaller than a pixel, photon mapping naturally turns into light tracing,and that’s when the method goes unbiased. After all ,blurring photons does nothing but give us a preview, and definetely brings error to the rest of picture.

Hi @shuzhan_yang,
and welcome to our forum!

Thanks for posting the stylish images.
SPPM is indeed a valuable addition to our standard path-tracer, enabling rendering of strong caustics. It has it’s use cases and, as you stated correctly, is a biased method (the blurring from the density estimation) but importantly consistent (it converges to the correct solution eliminating the bias over time).