CHALLENGE - Show your skills (scene included)

[LudvikKoutny]

Hmm... sorry it does not work for you. At least i tried

Chairs are darker probably because i do not use 100% white refraction, and diffuse is set to black. Also, exposure of my entire image is slightly lower than yours.

As for Irradiance Cache, it is not necessarily too low. You will not lose much detail as long as your scene is flooded with direct area lights from multiple angles, like in this case

[cecofuli]

Recon442, do not get me wrong. Your setup is a PERFECT ratio quality/speed. 20 min is awesome rendering time. Try to think to do an animation. 10 min with DOF in post! Very good!
But, for example, look at the bushes outside, the chairs, the glasses, or the lighting solution. It's good, but not the best we can achieve =) Thank for your time, really!

[Morbid Angel]

Im wondering why the whole image looks darker then the one posted in the first post. With reflection interpolation though you lose detail in reflections, I guess it does give you good speed but I'm not sure if that's the quality he is after.

[Morbid Angel]

Recon442, do not get me wrong. Your setup is a PERFECT ratio quality/speed. 20 min is awesome rendering time. Try to think to do an animation. 10 min with DOF in post! Very good!
But, for example, look at the bushes outside, the chairs, the glasses, or the lighting solution. It's good, but not the best we can achieve =) Thank for your time, really!

btw interpolated glossies will not work well with animation. Not certain about camera move only, but its sure to flicker in every other case.

[LarsSonparsson]

Im wondering why the whole image looks darker then the one posted in the first post. With reflection interpolation though you lose detail in reflections, I guess it does give you good speed but I'm not sure if that's the quality he is after.

I used interpolation only in materials that had glossiness set 0.3 to 0.5. They already had lost the details - in away.
But I dont really like to use it. When starting to compromise reflection, shadow and gi quality and loosing the dof and making the whole scene darker, there isn't any point tweaking anymore. We have already lost the quality we were looking for. This is hard scene to render, but 2 hours and still noisy.

[LudvikKoutny]

Im wondering why the whole image looks darker then the one posted in the first post. With reflection interpolation though you lose detail in reflections, I guess it does give you good speed but I'm not sure if that's the quality he is after.

Because i redone whole scene from scratch, lighting and materials, so exposure may be off, it's all just a matter of raising exposure or lights intensity.

And i do not use interpolated reflections, only caching reflections with GI in case of one material in the scene. And yes, this setup would probably not hold in animation too well, but that was not request in this scene. I usually differ scenarios and then base my setup on them. It does not make sense to for example render hires print still with animation setup, if the only difference it makes in the end is if it flickers or not.

As for the quality, as i already said, those things that look different look just different, not worse, with some additional time it should not be problem to match your output almost perfectly without any rendertime difference. I redone all materials from scratch, so of course there are deviations. My point was a proof of concept when it comes to rendertime.

[LudvikKoutny]

It is very important to be able to differ what are effects of appearance, and what are effects of quality. The fact my scene has slightly different appearance does not mean quality is lower, that is just a matter of perception of an individual.

Cecofuli for example says my foliage behind windows does not look good. Well, it looks different, but it does not look worse (especially because his foliage has almost no translucency, while my foliage has a bit of it, so i could say my trees are even more realistic :P ).

Non the less, it took only very little effort to get the scene looking a lot closer to Cecofuli's original without altering rendering performance in any way. If anything, i would dare to say my chairs have slightly more highlights as i use global trace depth of 8.



This took about 17 and half minutes on my Dual Xeon at work, so it should take about 35ish minutes on regular i7 2600k.

Cecofuli, if you want, i can send you this updated scene which looks closer to your original

[mienda]

a little setup with cecofuli lights and camera.
The same point view but with a new exposure.( shutter speed at 200)
Dual xeon 2.93
1280*800
AA 1-24 clr tresh 0.01_ Brute force subd 60_noise treshold 0.01
Light subd 24 and some material at 50 sub

The render could be less grainy and faster with some more test i think ( more subd on the black wall for ex)

cheers

mienda

[LarsSonparsson]

Wow! That looks really good.
Exposuring the scene better makes it easier to render, I think.
Also, I like it that way better.

[mienda]

A new version with more subd and with a corrected glass reflection !
a little bit longer but much better i think.

regard's
mienda

[jacksc02]

Cecofuli - can I have a go?

[cecofuli]

mienda, this is great quality/time!
If you have 1h30 min with dual Xeon, with our PC I can raise near 2 hours.
Some little observations: as you can see, again, my scene is a little bit more "vivid", with better (not better, more) highlight everywhere.
But I think cause you changed some lights intensity / camera parameters / glossy values.
BTW, can you share your setting. At the end of this nice thread, I would like to share with us the final scene, with the best setup we can achieve.
PS: if you can, post a PNG file. Sometimes JPG compression hides the noise.

[^Lele^]

Well, i gave the scene a whirl, and by doing so i think i lost the plot.

I do not understand what the thread is about: proper setting up of a scene for LWF rendering, or benchmarking what VRay's performance is when it's set up so to skew the light transport equation, and asked to trace an unnecessary number of (mostly bad) rays.
I have the scene down to a couple of hours (maybe 2:30. playing FM2013 while rendering doesn't help benchmarking), on an amd fx8350 (8x4ghz, all stock), through purely proper sampling (correct subdivs for materials and lights), and acceleration of the glossies through the Lightcache.
As it is, it's good to be used up to two stops brighter (four times as bright as the render) without visible noise (despite the LC caching it holds perfectly also in animation.).

However, the scene isn't set up as per what common practices would dictate, so i find it quite a sterile exercise to try and squeeze performance under these colormapping conditions.
The scene is surely a "difficult" one, however, 5 bounces of 0.3 glossy reflections on the white wall paint seem unnecessary, for an effect that under those conditions is much like a GI bounce (mayhaps with a slightly different hue), and for which the contribution for the first bounce is around an eight of the diffuse contribution (and an eight of an eight on the second, and so on).
Surely one can part from a 3rd, 4th, or 5th bounce of glossies before the render starts (the third bounce multiplier would be at around 0.001953.), instead of leaving VRay to try and trace a gazillion rays that get polluted as they travel across the scene (as the lighting equation is not linear and values don't behave as expected), generating nigh-uncleanable noise ?
Same goes for lights: with all those glossies around, what's the point of having recessed lamps visible to reflections?
Surely, in this case, one's looking for the light FLUX reflection (ie. the direct lighting on the wall), rather than the light FIXTURE.
However, as they are now, the superbright light fixture direct reflection gets picked up by the wall paint, and spreads noise which then requires massive sampling to clean.
That also applies to the sun.

It isn't the translucent material being slow, either.
It's that the base leaves material is opacity-mapped, and has five bounces of glossy reflections.
With translucency on, that doubles the calculations.
Again, try unticking "trace reflections" on the leaves, and ideally, spend some quality time generating geometry leaves: you're anyways saving out proxies, so geo sizes are no issue, while you'd give vray nice, opaque geo to begin with.


Not being under normalised, linear conditions means that the background brightness is arbitrary, and exposure chases a ghost trying to balance indoors and outdoors (ie. it seems to me the Bkg is composited in camera. Alpha+Post would be your friend, here).
The same applies for materials' albedos: being arbitrary, there is no way of figuring out how to make a real material look identical under these color-mapping conditions, aside from eyeballing it (which, allow me, is a bit last decade, and a tad.).

In other words, rather than trying to squeeze performance out of a scene which isn't quite rightly set up, i'd first make sure my scene scale is right, BOTH geometric AND lighting-wise (ie. radiometrically linearised).
Then, i'd render elements, and composite color corrections on individual ones later, instead of trying to wrestle indoors, outdoors and an arbitrary plate in camera, at 4 hours a render.

To make the scene linear is overly simple, with a couple of lines of scripting.
However, the look IN CAMERA changes dramatically, as the setup unhides the unbalances in lights and materials setup.
Of course, there's quite some serious speed to be had, sacrificing the effects which will be perceptually negligible, along with linearising the scene, and leveraging specialised rays, rather than generic ones.

However, somehow i don't think this was the original thread theme, or was it?

[mienda]

hi,

This is a new test with all cecofuli's materialsl and lights setup. I only change the exposure of the camera( shutter speed at 200) and the subdivisons of the materials. a little longer but more much near from the original.
This is not an easy scene, very bright lights and a lot of very glossy reflection!
The only way i find to render with less dots and noise was to check "use light cache with glossy ray" . it is very important and faster but it is less accurate ...without it an with your high glossiness ( 0.3 for example on the black wall) i think it is impossible to render fast and smooth

Anyway, i managed the scene element by element.
The subdivision of the shaders are defined by the glossiness with a range from 16 to 200 ( no glossiness to full glossiness).
Another important point to reduce noise is the use of filtering ( in my case vrayLancozosFilter at 1., it takes times to render but it helps a lot to mix all the elements together.
The lights are the same ( except the poral parameters setuped on simple)
The window's glass are corrected( delete the bad face)
This is my setup:

AA 1-24 Clr tersh 0.007
brute force subd 70
LC 2000_ sample size 0.02_ scale Screen_ pre filter (10)_Use light cache for glossy ray

DMC sampler Adaptive amout 0.85
Noise treshold 0.006
the rest by defaut

Hope it helps
Cheers

mienda

[cecofuli]

mienda,
did you change something in the leaves shaders (both trees and bushes)
they looks more green.