Vray - also a fast one button render engine?

[osuire]

Vray - also a fast one button render engine?

Being a newcomer to V-ray, I greatly appreciate to use predefined settings like these.
Compared to some other settings I found on this newsgroup, Micha's are incredibly fast : 10 minutes compared to 7 hours, for comparable results.
Moreover, I definitely think there should be an option to switch the interface to a "Beginners mode", where the user would choose only between a small set of pre-defined render profiles such as :
-Interior / Caustics (slow)
-Interior / No Caustics (fast)
-Exterior / Caustics (slow)
-Exterior / No Caustics (fast)

A simple slider called "Quality" LOWER<-- // -->HIGHER would allow the user to control the quality of the output very simply.
Now this may seem ridiculous to die-hard users, but I'm pretty sure this would please those of us who don't give a damn about the "normal threshold of the irradiance map", whatever that means...

Personally, I would much more fancy a physical sky and a proper "sun" with geographic settings and all that.

Cheers !

--
Olivier

[Micha_cg]

Vray - also a fast one button render engine?

[merlik585]

Vray - also a fast one button render engine?

[quote="Micha"]Best you do two independent test first.


At dual core engines you must set the bucket size at half image size.

So I guess for quad core I have to set it to 1/4 of the image size ?

This method is useful for test only, because the progressiv preview is not working (Vlado must make, that the buckets are updated every minute or so).

Ah!! that I did not know


Your visopt looks ok. I'm not sure you need the default lights for your scene.

The visopt I sent you were taken from your post on the chaos NG,may be I left the default lights on by error.
Thanks for your time on this.


Renee

[dalomar_cg]

Vray - also a fast one button render engine?

So, I've been keeping an eye on this for a few days now and its a very intersting concept. At least my understanding of what your going for is some sort of ppt kind of thing but using LC, IM, or QMC. Maybe if you use qmc for both primary and secondary bounces you don't have to worry about waiting for any precalculations, and you could literally kill the image whenever you want.

So here's my suggested options settings
Both bounces= QMC
Adaptive amount=1
Noise threshold=0
Adaptive QMC min/max=1/50 or 1/100 whichever you prefer
Set buckets to fit resolution depending on # of processors

In the few quick tests I've done this works well with both caustics and dof. At this point I think you may have to set higher subdiv on dof as there is a bit of noise in the blurry parts of the image.

There are two main disadvantages to this method right now. First of all is the high memory usage. I'm not sure if there's really a way around it right now, but maybe somebody (Vlado ) has some ideas. The other disadvantage is that the viewport doesn't refresh, which makes it hard to determine whether the quality is sufficient at any given point. This could probably be resolved by retrieving the sample information from the buckets at a given time interval that could be set by the user (you can do it Vlado )

I think this is definitely a very interesting concept to make vray work in this way. I'm not sure exactly how useful this will be, but I definitely think that somebody will find the "set it and forget it" approach useful. I will post some images and visopts settings tommorrow.

Edit:
So here is the first image, which is all default settings with IR/qmc for bounces. The caustics are precalculated so they don't play into the final render time.


This image is done with the progressive qmc settings above and also with the same precalculated caustics. I specifically stopped the image at almost the same time so that the quality of the images could be compared for the same amount of render time.


Just for comparison here is a PPT version. The qmc sampler settings are back to the default values. Again the caustics are precalculated and I stopped it at about the same time as the others

[merlik585]

Vray - also a fast one button render engine?

Very useful comparison.Seeing this,it reinforces my opinion,the first one is the best,and for me there is no real advantages in using the other method.As you mention I don't know how useful it could be,but we never know.For now I will stick with my settings,this is what work best for me.

Renee

[joe_b]

Vray - also a fast one button render engine?

I definitely think there should be an option to switch the interface to a "Beginners mode", where the user would choose only between a small set of pre-defined render profiles such as :
-Interior / Caustics (slow)
-Interior / No Caustics (fast)
-Exterior / Caustics (slow)
-Exterior / No Caustics (fast)

That is the general idea we're going to be working toward in the future. Basically we'll be hiding most of the options from the user ( except for the advanced ) and making things more slider oriented...low quality -> high quality. These UI changes won't make it to Rhino by the service release, but for our next major version we will be doing some heavy UI reworking with the idea of Beginner/Intermediate/Advanced users

Personally, I would much more fancy a physical sky and a proper "sun" with geographic settings and all that.

We are working on figuring out the geographic settings stuff for the V-Ray physical sun.

[Travis Serio_cg]

Vray - also a fast one button render engine?

That sounds great Joe. I agree with your theory and concept. I find that most CAD operators spend more time in learning the art of modeling and rendering is given much less focus. In fact, the majority of cad operators know little to nothing about the mechanics of rendering period. I think basic simple settings combined with stunning material libraries and environments puts great renders in the hands of the people that wouldnt otherwise render. The easier it is to render the easier it is to sell copies, and everyone loves selling more copies

[Micha_cg]

Vray - also a fast one button render engine?

Dalomar, thank you for testing the method. I see the biggest advantages, if a LC prepass is used. The LC provide some useful advantages:
- infinite GI bounce calculation
- sample size (first short cut)
- filter option (secon short cut)

The two short cuts allow, that all not direct visible GI bounces can be calculated faster, not so high detailed. Best you try it with a more complex scene like an interior.

PPT (curtain with SSS)

PQMC based on QMC+LC

PQMC based on QMC+QMC

Universal IM based on LC

Universal IM based on LC - progressiv rendered and stoped after 12min


The most speed enhancement we get, if we use all GI methods with high approximations - LC + IM - as base for the final progressiv process.

I hope, Vlado can modify the PPT mode so, that it can be used like the PQMC and so the memory/update problem is solved fast.

[Jean-Paul Paloux_cg]

Vray - also a fast one button render engine?

micha what are the setting for the last scene of 16 min! ?
you have the visopt file ? it's including light cache calculation ?

[Micha_cg]

Vray - also a fast one button render engine?

So I guess for quad core I have to set it to 1/4 of the image size ?

Yes, this is the current experimental workflow.

[andreas]

Vray - also a fast one button render engine?

Hello Micha,

I just returned from 1 month of vray absence to see your post, which I find very interesting, especially the ?fire and forget? aspect of it.
Seeing that your last post is dated Jan 08 I wonder if there is any new development on that issue, have you received any reaction from Vlado yet concerning your ideas about the update feature of the frame Buffer?

I think you are really brilliant in using existing techniques and methods to create new ways of render approaches! Keep up the good work.

Best regards

Andy

[flipside]

Vray - also a fast one button render engine?

Note that these settings are resolution dependant. Ir map @ -3/0 on 800px image will give good GI quality, but use the same on a 5000px image and GI quality will be extreeeeeeeeemly good (and take a lot of time...)

What is PQMC?

[dalomar_cg]

Vray - also a fast one button render engine?

PQMC is supposed to stand for progressive QMC.

The idea was to have a ppt like render that would be faster and produce better results than ppt. The idea was then to use qmc's ability to show the calculated samples before the bucket actually is finished. By setting the bucket size to something that would basically cover the whole image it would be possible to set the quality to something quite high, but kill the render whenever you feel like.

[Micha_cg]

Vray - also a fast one button render engine?

So, I'm back. I have overseen, that the forum january blackout was finished after 2 days and I was waiting for a news at the home page and waiting ... .

Vlado write me about the PQMC per email:

Hello Micha,

The method that you describe is quite interesting. In fact, I have been thinking of implementing something similar.

Basically this works like the "universal settings" method, but the progressive sampling is applied to the entire image, rather than each bucket in turn.

So we will definitely work on that.

Best regards,
Vlado

I hope he can implement it so soon as possible.

Back to my experiments. I have done many more tests in the last weeks. I will try to find a conclusion for now, so we can discuss parts of them:

*(1) progressiv rendering of the whole image based on QMC image sampling (PQMC)
... this idea is general independent from GI and the goal is, that user must not touch any subdiv setting of blury values

*(2) the PQMC method can be used with useful GI combinations like IM+LC or QMC+LC.
... QMC+LC provide a high quality for GI and blury effects without interpolation errors. This method is faster than pure PPT. One LC prepass is necessary befor the progressiv final rendering start.
... IM+LC provide all advantages of biased renderings: low noise and very fast rendering. Two prepasses befor progessiv subdiv sampling is necessary - LC and IM. The IM pass need some knowledge about the usage.

*(3) General full adaptive sampling for non GI effects. I have the feeling advanced users of vray don't like to setup Vray in this mode, because this metod could be slower than manual setup.
I'm not sure. I think, this method must not be slower than manual setup of every subdiv value. I think, a specific image noiselevel need a specific subdiv level. A manual setup must not be faster for the same quality, because the needed samples must be calculated too. But it's not easy to find the right manual subdiv value, because it is independent from the intesity of bluryness.
An extrem example of the disadvantages of non full adaptive sampling show me my velociraptor scene with DOF enable (subdiv . If I hide the wood dino, an render the scene in full adaptive mode than the render time is 11s. If I set adaptive amount 0.85 and QMC 1-4 (befor 1 / 1-100) than the render time is 22s, because the DOF subdivs are not 100% adaptive anymore and the rendering slow down. The DOF effect is not visible in the scene, because no object (exept the ground plane) are there, but the manual DOF subdivs are sampling the ground plane to much, more than necessary for the visible effect. The full adaptive sampling take in acount the visible effect of a subdiv and calculate necessary samples only! You get what you need - a specific noise level and not a specific sampling level.

*(4)The influence of the full adaptive mode to the GI calcuation of the Irradiance map.
This was a shock. In some cases the IM dosn't work right with the full adaptive mode. At my user tutorials I recommend to use the full adaptive mode ala Vlados universal method, but I went a step further. Vlado use the full adaptive mode for GI QMC+LC. I thought, I could speed up the calculation and used IM+LC. Works fine often, but not allways. Problem: the IM calcualtion show ugly artefacts (link).
Why? What is the reason? The main idea of the full adaptive method is, that all subdivs values are determinated by a global noiselevel. Which subdivs are controlled? Blury reflection/refraction/DOF and IM samples. IM samples are noise? Not realy and this is the problem. A noiselevel of 0.01 can be good for the fine noise of reflection/refraction/DOF, but can be bad looking for splotchy IM samples. We have two kind of noise - high and low frequent - blury effects and GI effect. Low frequent noise need an other noiselevel as high frequent noise. A noiselevel of 0.02 produce an acceptable noisy look at reflections, but at the IM it's horrible (see attached test image).
This problem is not a problem of full adaptive mode only. My experience (and Wouter tell about it too) show me, that the manual setup (adaptive amount 0.85) show this cross dependence problems to, if a noise level higher than 0.01 (0.02 or more) is used.
What could be a solution? I think an independent adaptive control for the IM could help. My dream is to setup the IM in independent full adaptive mode, so that the user can set a splotchiness level independent from the noislevel of the final beauty pass.

(1) and (2) can be used, if Vlado implement, that the QMC can be set in progressiv mode for the whole image with continous updated framebuffer. So, I hope it's only a question of time to get it.
Point (3) should show that full adaptive should help to get good results in most cases, maybe faster than per manual setup. Biggest problem for a universal setup is point (4).



Here a speed test, what a full adaptive IM could do:
-download the scene
- render the scene with "don't render final" and noise level 0.002
- note render time t1
- save IM cache
- set IM to load the cache
- set noise level 0.02
- disable "don't render final image"
- render final pass and note time t2
- image looks perfect

Both times added give the full time of a high quality full adaptive IM based image. Now try get this result with manual setup in one pass with adaptive amount 0.85. My t1+t2 rendertime is lower than all my manual test same quality.

PS: I spend to much time for that all. I'm happy about every feedback and help.

[Gijs]

Re: Vray - also a fast one button render engine?

Hi Micha,

thanks for all your efforts diving into this and doing all the research. I know this is very time consuming and it looks you worked hard on this.
Can you fix the links to the topics? they are referring to the old phpBB2. Also the link to the test file has become an internal link, you might have forgotten to put http:// in front of it.

I just downloaded the file and ran the test as you suggested. For this simple scene I can say it works pretty well. We might need one or a couple of difficult benchmark scenes to test these settings on, render it with your settings and then try to outperform it with manual settings in order to see if it is truly universal.