guibou wrote:- Are you using SPPM for your work and why ?
Not at the moment, but I will take it into consideration in the future.
guibou wrote:- If no, why ?
I have problems with the robustness of the algorithm. Not all, but some problems show up that relate to the regular photon mapping. You set the scene up, you start rendering and hope for the best. Theoretically, there is no QoS regarding the correctness of the result in a reasonable amount of time. It very well may be that you have to restart a render with different settings after an hour because of a glitch, while I am really used to the luxury of the "fire and forget" concept of metropolis light transport. It also seems to have difficulties with more complicated materials (such as glossy things, but a gold/glass mixture also gave it a hard time).
guibou wrote:- What kind of work are you doing with it. Is this hobby, are you in a small art company, are you part of a big company ?
My fiancée is a 2D graphic designer and I help her making better visualizations of her work by using 3D rendering. Imagine business cards, wine label designs and these sorts of things put in a 3D environment.
guibou wrote:- What are for you the pro and cons of the method ?
- An important con is the high memory usage, though it should also be noted that the photon shooting phase does not separate direct/indirect/caustic/radiance photons anymore, which is a blessing. I remember the photon mapping times when I had to wait loads to be lucky and get enough caustic photons. No more of that!
- This particular implementation lacks the volumetric scattering calculations, but the algorithm itself is obviously capable of it. It would be a great addition.
- I didn't like the parameter tuning of the starting radius and alpha. Theoretically, in the limit (t->\inf), pretty much all setups would work fine, but one setting could make a good to go render on a diffuse scene in 3 minutes, where an other would leave a splotch here and there which doesn't really seem to disappear over time. The fact that it will converge to the correct solution in the limit (an infinite amount of time) doesn't give a lot of comfort here.
- Another pro is that it can handle horribly difficult light transport situations well, and a rough preview of the final render will almost always appear within minutes.
- I don't know of any SPPM implementations around that would be as usable and speedy as this (and yes, I know the Hachisuka reference implementation and also Mitsuba). Great work.
guibou wrote:- What can be improved ?
I think making it more robust would help the most. This is something regarding the algorithm itself, not this particular implementation, and it would make a great research topic.