[Bf-cycles] BiDir on Cycles...

Brecht Van Lommel brechtvanlommel at pandora.be
Thu Nov 24 13:14:18 CET 2011

My point wasn't that these features are not useful for architectural
rendering, they are. It's that the type of lighting that bidir or MLT
can render faster than path tracing (caustics, very indirect
lighting), are off the table anyway for rendering frames in an open
movie project. I just don't see how we can get these rendered in a
reasonable time regardless of the algorithm.

'Basic' things like direct lighting with many (area) lights, one/two
diffuse bounces, glossy reflection of environment maps, ... already
present a major challenge. They require improvements to algorithms,
things like decoupling aa/diffuse/glossy samples, solid angle light
sampling, adaptive sampling and/or caching and image filtering with
ray differentials based on the BSDF.

>From that point of view, bidir or MLT become low priority. Someone
else is welcome to work on them of course, but I can't justify
prioritizing them at the moment.


On Thu, Nov 24, 2011 at 10:12 AM, ertuqueque <ertuqueque at gmail.com> wrote:
> Half of my work is architectural rendering... The other half is making TV
> commercials, VFX and general audiovisual (3D) design projects...
> Actually, all those stuff you mention (render passes, render layers, motion
> blur, displacement, subdivision, caching...), are equally important for
> architectural rendering. Is a common error to assume that for architectural
> work, Lux (or any other physically based / spectral renderer) is a better
> suit for the task and you can do without all stuff you mentioned... In all
> the architectural projects I've made, I have used A LOT features like Render
> Passes, Layers, Compositing... I've even used the faulty micropolygon
> displacement from the Sintel's Render branch once...
> In case you are curious:
> (My lastest architectural project, Blender Internal)
> http://vimeo.com/25014763
> (This bellow is the one with the micropolygon displacement and GI from the
> Render Branch)
> http://vimeo.com/15709575
> And the architectural project I'm working on right now (my first made with
> Cycles), is also going to be animated, a walkthrough of about 1:30 minutes
> at HD 720p...
> Maybe my work on the bmps has made people thing that I don't like to do
> optimization on my projects...
> No matter the type of project I do... Architectural, character animated TV
> commercial, VFX... My approach is the same.
> This project has a buttload of optimization going on... Caustics are
> disabled, Diffuse Bounces is set to 3, Glossy to 5 and Transmission to 8.
> Most glass objects have shadow casting disabled, there's a special node
> setup in the environment to keep light from entering inside the building. In
> the windows there's another node setup so they look as glass but they emit
> light almost like a "light portal". There are several objects in the scene
> that are not visible to the camera and only cast shadows to avoid some
> fireflies caused by lights very close to the ceiling (thank you very much
> for the "Ray Visibility" panel and "Light Path" node btw)... I can keep
> going on and on. I can even send you privately the .blend scene for testing
> if you want...
> An (improved) image to show the result:
> http://s5.postimage.org/i0zfmffh3/image.jpg
> But there's just so much you can do with optimization "tricks"...
> What I'm trying to say is... At least talking from my experience, an image
> is an image. Being architectural, motion graphics, a scene from Elephant's
> Dream, BBB, Sintel or a Suzanne on a plane, the approach is always the same.
> I'm not trying to change your priorities here, don't get me wrong, you know
> better than me about what's needed. But imho, improvements on the Path
> Tracing algorithm should be right after those stuff you mention (maybe even
> as a parallel development)... It would be even better if another hardcore
> developer helped you with that. I know implementing a CPU/GPU BiDir Path
> Tracing isn't a trivial job.
> Anyway, sorry for the long and maybe patronizing reply... I just wanted to
> prove my point, I have to admit that it bothers me a lot when people
> separate 3D imagery into categories as if they require radically different
> approaches.
> Greets and thanks for your hard work.
> tuqueque.
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles

More information about the Bf-cycles mailing list