closedspaces

[lacilaci]

So here's some random spare-time stuff:D Almost all have default corona setup. Of course not to be considered as a real comparison between different illumination modes... Only top row was rendered with a 30 min. limit and only the two rightmost pics have glossy refl floor.

All lights have a one face emitter enclosed with twosided (oneface) material except the one marked in bottom right image which has thick geometry (one sided solid material).

Like I said. It's not a "real" comparison. However I was stunned by the result of IR+PM it's quality and speed. I failed to get it nicer in the reflection nothing was helping to get it cleaner (to fix photons visible in reflecitons). Also PPM seems like a great thing but the longer it renders the less efective it is even with changing alpha reduction and photons per iteration settings(at least in my observation).

Any tips how to get nice reflections with IR+PM combination?

[Paul Jones]

IR as a primary GI engine is not complete as yet - use either:-

• PT as primary, HD cache as sec
• PT as pri and sec
• Bi Dir
• VCM

[Ondra]

The compression is totally killing the comparison :/

You can also use this for comparing images: https://dl.dropbox.com/u/5426917/comparison.html - just download and edit the file, replacing the images URL with your own (displayed text is the alt tag)

[lacilaci]

The compression is totally killing the comparison :/

You can also use this for comparing images: https://dl.dropbox.com/u/5426917/comparison.html - just download and edit the file, replacing the images URL with your own (displayed text is the alt tag)

I'll take a look at it tomorrow I'm tired I've been looking at that stupid scene whole day :D...
However I'd still like to know if there is/will be a way to optimize how photon mapping (as secondary GI) looks in reflection. Cause otherwise it seems like an awesome and fast solution.

[Ondra]

you can use VCM without connection, that is called bidirectional photon mapping. I could also add some better heuristic for deciding when to shoot rays/when to interpolate, but overall photon mapping sucks - just try to add small details, or light the scene with environment ;). So it is not a viable way to develop Corona in the future.

[tomasd]

Useless note: All algorithms are "less effective" the longer they render. To reduce noise 2x, you need 4x longer time. But yeah, PPM is even worse at this.

[mjkl]

nice comparasion