Progress Report

[Roy Hirshkowitz]

I'm working mainly on Path Tracer stuff-- here's one of Peter's where I tagged the windows as caustic and stuck a rubble bump map on them.  This only ran for about 15 minutes, so it still has a little ways to go, but you get the idea.  Caustics lighting is now validating much better:

 

My daylight source experiment didn't go that well-- but the glossy stuff is working a little better. Ando's Church of Light provides a very difficult lighting environment.  No caustics-- ran overnight (still a few fireflies):

 

[Ros. Riha]

looks very good!

[Marc Chaumier]

Nice, Roy, could you post the same work, with the old calculation?

[Roy Hirshkowitz]

Not easily no-- however I did have the math wrong for the above example.  Here's what it looks like now (no fireflies that I can see, either.)

 

[Peter Milner]

How many passes did this one need to resolve so well?

[Roy Hirshkowitz]

Hard to say-- it ran overnight for > 8000 passes.

 

Here it is after three hours at 2243:

 

[Marc Chaumier]

Looks good, but the floor is strange... perhaps need it a little bump or less reflection. Nevertheless the illumination seems to be real (white halo). The sun is straight behind the wall, right in front of the observer, just a little bit on the left, true?

[Roy Hirshkowitz]

It's the same material as the pews in this example-- not correct.  Since I'm working on glossiness-- however-- both the concrete and the floor are helpful.  It is, like I said, a very difficult lighting environment.  Things that make it difficult include:

• Small unobstructed windows
• Large obstructed windows (there is a wall just outside the large windows which are mostly behind the viewpoint.)  This causes a large portion of the ambient light to be received via a diffuse bounce or two from the sky-- then through the glass.  Daylight sources do very poorly with these.
• Glossy surfaces-- these would have taken much longer to resolve using the old algorithms.  They resolve quickly now.  The stubborn part of this image is mostly the diffuse noise along the back wall.

[Marc Chaumier]

Last question, I know it's hard to define, but is it possible to give us an idea of the improvement ratio?

[Roy Hirshkowitz]

Not really.  The caustics stuff has improved quite a bit in terms of possible application (not necessarily performance).  The daylighting may have gotten a little better-- not the big performance gain I was hoping for-- but glossy surfaces should resolve quicker.