The información in this guide can be found in this hilo in the Yafray forum: enlace todo interior lighting, noise and difradius researching.
Tricks and tips for Photon Maping. Basically, you have four result-levels for Photons:
* type 1. Dots or blod cell circles: means you are looking up to few of them.
* type 2. Well covered but cloudy/splotchy (not smooth): looking up a tad todo.
Few.
* type 3. Smooth and detailed: just right.
* type 4. Over-smooth and totally smeared : looking up to Many / todo large Radius.
Basically, the Photon search estops when either the Radius is reached or the count is reached. So if you set it todo 100 Photons, 10 centímetros, it Will look within for the 100 nearest Photons within 10 cm.
* if the 100 nearest are within 1 centimeter, those 100 Will be used, none further than 1 centímetro Will be used.
* if the 100 nearest are within 10 centímetros, they Will all be used.
* if the 100 nearest are within 1m, only those within 10cm Will be used.
From this you can trivially understand that if you hit the number limit this acts as an adaptively variable Radius, the more Photons, the better the detail, the less the Photons, the more Radius is loked up in, however, if the count is todo high, the Radius is always hit first. So you Will always smear the result over the Radius. This is oversmooth (type 4 above), the trik is todo get todo type 3 just right, but the god thing is that when combined with FG, the accuracy ned is much less. The cloudy (type 2) solution works just fine with FG. A spotty/circly (type 1) solution wont work, a over-smoothed (type 4)
Solution works but can put light where it not supposed todo be, so with FG the Margin of error is large, my workflow is generally:
1. Turn of FG.
2. Lok at the Photons-only result.
3. Is it dotty? Increase Radius, or Photon count, until it estops being dotty.
4. Is it super-smeared, decrease Radius.
5. When neither dotty Nor super smeared (generally cloudy or nice), be happy.
6. Turn FG bak on.
7. Rejoice.
o god aproximation examples.
Its better if we dont consider dif radiusvalue alone, but a as a combination of both dif. Radiusand search. I mean, it is the number of Photons found (search) within a Radius (dif Radius), if you decrease search, there are less Photons todo mix therefore the likeliness todo average a diferent result than the neighbour patch increases, hence the more noise.
search 20
Dif Radius 1.0
Photon count 200.000
If you increase search, there are more Photons in the mix, therefore the likeliness todo average a diferent result than the neighbour patch decreases, hence the less noise.
search 150
Dif Radius 1.0
Photon count 200.000
Dif. Radius is the size of the Photon map patch. If we decrease the size, we are increasing the signal frequency, there are more patches therefore is a more precise Photonmap, but at the same time the noise produced by searchis increasing its frequency.
search 20
Dif Radius 0.1
Photon count 200.000
So better we use a god searchvalue. However, as we decrease dif Radius, there are some patches that arent getting enough Photons for searchto mix, hence that group of patches Will have got less luminance. An insuficiente Photon countwill introduce Low frequency noise again, Even on a high resolution Photon map.
search 150
Dif Radius 0.1
Photon count 200.000
Photon count should be increased as dif. Radius decreases:
search 150
Dif Radius 0.1
Photon count 2.000.000
150 search.
Now that a Photon map ready for Final Gather todo mix.