Yafaray, Indigo, (Luxrender) ...

[enricoceric]

Hi all,

Continuing my tests with the available render engines used with Blender, I tried an indoor scene based on a reference image I found in a “Chateau d’Ax” magazine. I tried to reproduce the lighting used on the advertising but I met a lot of difficulties to achieve the expected result. In the photo you may see the interior as well as the landscape through the windows, such a scene doesn’t allow the usage of Portals (for indigo and luxrender) and this has an impact on rendering time. Another difficulty was to reproduce the interior lighting based on the usage of several lamps and a overexposed sky without using sun, so after alot of tests I finally used a completely white HDRI image together with several mesh lights for Indigo and a single white color with several lamps for Yafaray to illuminate the scene, this situation is very hard to tweek.

So here are two images the first rendered with Yafaray and the second with Indigo, the Luxrender version will be added later on.

Some tecnical information :

Modelled with blender 2.48.1 Optimized

Textures from 3DTotalTexture DVD and http://www.cgtextures.com/
Trees from http://yorik.orgfree.com/greenhouse.html and http://graphics.uni-konstanz.de/plantslib/
The far landscape with trees is a bilboard with Alpha channel for Yafaray and a blend mask for indigo
The carpet has been made using particules reproducing 12000 bilboard meshes textured with a hair image. Again, I used alpha channel for transparency with Yafaray and a blend mask for Inidigo.

Some additional images made with Indigo only, some are older and the scene may have change since I made them. The stair and upstair are made based on several images I found in an architectural book (Techniques & architecture).

http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender4_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender5_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender6_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender7_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender8_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender9_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender10_3.jpg
http://www.myline.be/myline/3D/tests/testIndigoYafarayLuxrender11_3.jpg

Luxrender :
Last CVS version from 19/11
Metro + Bidir (Bidirectional is still in development and this is not an optimised version)
Lighting = White hdri map + mesh lights
Rendering time : 9h in 1200x840
Some postpro to remove few noise.

Yafaray : Version 287 Photon mapping, 32 path samples, AA=1, 10, 3 Lighting = Single white background color + Lamps Rendering time : 6h30 in 1500x1050 Some postpro to remove some noise

Edit:I replaced the previous version with this one rendered using Photon mapping instead of pathtracing

Indigo : Version 1.1.14 64 bits Metro + Bidir Lighting = White hdri map + mesh lights Rendering time : 51h !!! in 1500x1050 Some postpro to remove noise, even after 51h, this huge rendering time is mainly due to the lighting option (hdri + mesh lights), I'll try later on with sun only.

Yafaray

Sunsky background with sun emitter
Pathtracing

Indigo

Sunlight without exit portals
Metropolis+Bidir

Luxrender

Sunlight + exit portals
Metropolis + bidir

Indigo

Sunlight without exit portals
Metropolis + bidir

[Meta-Androcto]

wow,
There’s some interesting differences going on there.
I think Yafaray is less quality, but there are areas where it seems better.
For some reason the line of sight to outside looks more realistic.
Much more natural reflections I think. (but i’m no expert!)

The Indigo render overall is better.
The rear light shadows, the carpet, the eggs, but not the table…

hmm, by the looks of this,
Yafaray has better reflections & Indigo better shadows.
Interesting.

Oh yeah Thanks for the great work.

[Alvaro]

have you forgotten how to make good renders with Yaf(A)Ray, Enrico?

[enricoceric]

@Alvaro:Well, I should have missed something but this scene with that lighting setup is just the hell. I’m making some new renderings using sunlight instead, hope something much more sexy will come out

[organic]

I had considered something like this myself, but I don’t know the renderers well enough to be fair to them.

The difference in colouring is striking. I agree with Meta-Androcto about the view outside and the shadows, though in my opinion the Indigo shadows also look too deep(Could be my monitor).
The differences in the tabletop reflections are interesting too - though it mat not be a like for like comparison.

But what happened with the floor? The Yaf(a)ray image seems to be entirely missing the floor texture.

Really interesting series enrico, thanks again - will watch with interest.

[enricoceric]

Thanks Organic,

Indeed there is a problem with the bump on the floor, I had to increase it a lot to get it, Yafaray uses the Blender values and I had to push it from 0.5 to 3, don’t remember to have to set it so high with previous versions but I may be wrong. So I just rerendered this part and updated the image.

The difference for the top of the table is simply due to a material definition in Yafaray which produces to much reflection, I should also change it to get something else.

[data65]

Nice work! Without being told it is CG, no one would know. Just out of curiousity, could you do a render of that same scene with the internal renderer? I would like to see the differences.

[BbB]

I can’t really comment on Yafray, since I haven’t used it in a few generations. But both images strike me as really good and highly realistic. Indeed, the fact that Yafray’s output is so similar to that of an unbiased renderer seems to me to underline the renderer’s quality, Alvaro’s comment notwhithstanding.
The usual Enrico quality, in other words.

[Carbonflux]

Yet, can you prove Indigo is unbiased or what Yafray is going without open source? the point of unbiased is not to make the image look good, its to simulate photography which can come out looking very Bad. If you take a bad photo its not the cameras fault

I am not saying unbiased is ready yet, but it will be and given the new generations of multi core cpus coming out soon its going to be a lot more practical too.

Other then PBRT imo Luxrender is the only Proven unbiased rendering engine because you can view the source, its accuracy not quality that is the issue here imo.

Thats not to pick a fight of course, I have nothing but respect for everyones views and their work, but I do feel strongly about this, I am not sure if unbiased is fully understood by the community as a whole yet in that it keeps getting compared to ray tracing based on image quality which by definition a ray tracer is Faking, no different in many ways from converting what you learn in art school into a process, but for serious fine art work down the road imo ray tracing and even semi-unbiased misses things, in the end you should be able to take a photo of something and reproduce it in unbaised to the limit of our knowledge of physics.

Anyway, what I mean is unbiased should not be evaluated as a technology on if it “looks good” or “looks right” imho

My work might “suck,” but its my fault, not the rendering eng heh.

[thirdsense]

My work might “suck,” but its my fault, not the rendering eng heh.

hey, i like your selfcritique view of the workprocess. i think your are absolutely right. i wish more people would think like you do. but this has nothing to do with the software. even a photographer taking a picture of the “real” world will fake light etc. so realism is relative.

we need comparisons like this one. as an architect and a freelance doing architectural renderings i need to create “nice” pictures in a reasonable time, with reasonable effort. and i really appriciate to get a feeling of what can be done with which software.

concerning art, it might be a different thing. also i think that especially in artwork you need even more the possibility to experiment with the “viewing experience” and physical “wrong” images. realism is just one small part of the world of artwork imo. some of the best architectural renderings i have seen, really don’t have anything to do with realism. but they transport a strong mood and ambience in their fake way of showing the things.

sorry, i had to reply to your comment, even if it is maybe a bit offtopic here:)

regards
thirdsense

[BbB]

Well, I guess everyone’s appreciation of biased and unbiased depends on what we’re trying to achieve in the end. For the past two years, I’ve been using almost exclusively Indigo - not because it was physically correct, but because it allowed me to obtain the kind of images I wanted (mainly interiors) more easily and with better results than Yafray, which I had been using until then (although I would put this down to my own ineptitude rather than any shortcomings of Yafray, which is certainly a great engine as Enrico demonstrates).
I couldn’t actually care less about the “physical” aspect of things, provided the engine was generating the kind of images I needed without me getting bogged down in tweaking thousands of settings. I also do a lot of post-prod on my images to make them look more realistic or pleasing and I have no problem with this being un-physical or inaccurate or fake… As a tech idiot, I care only about the end result, not so much about what’s under the bonnet, which I don’t understand most of the time anyway, or whether it’s open source or not. Makes no difference to me.
I’ve now migrated to Vray for exactly the same reason: it gives me the images I want - and it is damn fast too. I’ll still keep using Indigo when it makes sense, and it has added a lot of really interesting features lately. But I think of all these engines as tools - completely un-emotionally and only looking at what they can achieve VS what I need.
Sorry for the long rambling, was just my two cents…

PS: Carbonflux: There’s one point where I totally agree with you: an unbiased engine is the closest you can get right now to taking a photo. But as we all know, it’s extremely easy to produce crap with even the best single-lens reflex or large-format camera. Which is why I never understood Maxwell’s slogan: “As easy as taking a photo”. Photography is where I come from and I always thought there was nothing as difficult as taking a good photo…

[Alain]

Very interesting comparison !
Are you able to compare it with V-Ray as well ? Maybe BbB can help you with that, he’s an expert in using V-Ray in Blender

Kind regards.
Alain

[Carbonflux]

Thanks for responding so gracefully, I was a tiny bit worried I was picking a fight, I think what I meant was understood perfectly, its really a open issue imo, but yes off topic, I do have to add one thing tho about “looking right” vs art, on some of my human characters I distort body parts, for example on one I make the hands and feet just a tiny bit bigger, its a reference to how blind people draw themselves which is a message i want to send but normally the response is “looks like some of your proportions are off” which would not happen with a photo clearly, so this is the problem with realism and CG imo. I took 3 years of anatomy at a one of the better art schools in the US so I figure that gives me the right to play around with non-perfect figurative work, even tho we are making progress I feel there is still to much of a “heroic” standard in terms of CG figures. If you draw from life etc you will realse real people are Way more distorted then you could get away with in CG.

[BbB]

I don’t know why but it reminds me of the VFX team on “Cloverfield” who said they struggled with the sequence when the statue of liberty’s head lands in midtown Manhattan. They said the head “felt” too small to anyone who was watching the sequence, so they had to make it twice the real-life size for it to “feel” right. Now that’s definitely off topic, but I think that’s what we all do when we tweak proportions or adjust our “photorealistic” renders in post to give them this extra something…

[radiance]

Hey,

Enrico, i’m wondering why you mention you can’t use portals in this scenario.
Why is that ?

If it’s because of the glass sitting in the windows, you can use portals as long as you use ‘architectural’ glass. (it’s an option in lux in the glass material, and i think indigo has it too.)

Radiance

[BbB]

I think he means portals exclude all geometry that is outside - i.e. “behind” the portals… The exterior would just render as sky or white…

[radiance]

Ahh

I though he used an exr map

Radiance

PS: Would be cool to see a comparison between yafray+luxrender afterwards, with lux configured biased/equal (photon mapping integrator). yafray will probably dominate as we’ve let that area slide for a while… nevertheless it would be handy to see the difference.

[enricoceric]

@Data65: I’m not sure I will do the test with the internal renderer, mainly because this would oblige me to brake the settings I use for Yafaray as it inherits values like bump, alpha, … and expected values for BI may be different.

@Radiance: Indeed, as BbB said, there is a problem to render external geometry using portals, at least with Indigo, do you mean that using architectural glass (as I do) allows to render external geometry together with portals ? By the way, this problem should only occures with sunlight as, if I’m not wrong, portals are not applicable with HDRI map or mesh lights, and indeed I used hdri map in this render but not for the other I’m rendering.

@the others: feel free to make phylosofical comments

[Rinne]

Path Tracing is not meant for interior scenes and mostly the cause of the noise in the Yaf(a)ray render…

use Photon Mapping, increase Photons, FG Samples, as well as FG bounces and increase the search value up to 1000 (will remove noise, but also cause less realistic lighting, you need to find a balance here)

Use few AA Passes (generally 1 should do) and 4-8 Samples…, higher for better quality…

Either use gauss (smoother, less noise) or Mitchell (sharper, more details) AA, but don’t use Box!

I’d recommend to use a small, fast renderable test scene for both renderers to tweak their settings…

great scene btw

[enricoceric]

@Rinne: I did several additional tests using photon mapping and indeed it seems to be fast and it generates less noise. By the way I completely messed up with the AA settings, I checked other scenes made with previous Yafaray versions and I noticed that I inverted the two last AA values, which dramatically increased the rendering time. So this means that I should recalculate at least this scene. I will soon post another scene rendering with sunlight and IMO a better result.