Problem of cloud noises on surfaces

[Fredd]

I have a problem of cloud noises on surfaces (Watch pictures). I don't know which settings I have to change to solve this problem. Thanks a lot for your help.

[Micha_cg]

Problem of cloud noises on surfaces

Looks like Irradiance map artefacts. I would increase the sample cound at 64 or 128 or 256. Also set color threshold at 0.2 befor.

[flipside]

Problem of cloud noises on surfaces

Micha, are you aware that 50 hsp actually means 50*50? I have never needed the hsp subdivs higher than 75. 250 subdivs will render 25 times slower than 50...

In interiors a lot also depends on your secondary engine, which is probabley left at QMC here. Try the lightcache with default settings to see if it helps.

But better post a screenshot of ALL your settings.

[Micha_cg]

Problem of cloud noises on surfaces

Wouter, I don't mean the subdivs, I mean the samples.

Here a screenshot of my favourit default settings. Fred, if you like you can jump to secondary engine -> QMC. This render slower, but you get all details without artefacts. The noisethreshold of the QMC sampler control the quality: 0.02 preview or high resolution image - 0.01 fine noise only - 0.005 approx. noisefree.
Also I set the framebuffer at "sRGB" and use a light s-curve at the color curve correction for a little bit more contrast.

[flipside]

Problem of cloud noises on surfaces

Ok, sorry Micha But also there, watch out you don't blur your GI too much, because the 'samples' is actually 'interpolation samples', meaning that you specify how many surrounding samples are taken into account for interpolation. So if you use 250, the 250 nearest samples are all averaged to give you the illumination in the currently calculated sample. In other words, it will look smooth but undetailed.

At least it's the case in max, I assume it will be similar here.

Another question, if you use the srgb button, do you correct your textures so they don't look washed out?

[Micha_cg]

Problem of cloud noises on surfaces

Ok, sorry Micha But also there, watch out you don't blur your GI too much, because the 'samples' is actually 'interpolation samples', meaning that you specify how many surrounding samples are taken into account for interpolation. So if you use 250, the 250 nearest samples are all averaged to give you the illumination in the currently calculated sample. In other words, it will look smooth but undetailed.
....

I'm quite happy with the results. If I have the feeling, I need more details, than I use -2/0. You recommend subdivs, which values do you prefer?

Another question, if you use the srgb button, do you correct your textures so they don't look washed out?

In my last render tests I have started to use the 0.45 gamma. I'm a little frustrated. So much trouble to get the right result. And I can not use the Rhino object properties textures anymore. If I use a scene with 20 lowpoly people and some cars, than I must create 20 materials for a simple diffuse textured color.

In the past I have thought, if I use gamma 1 and my textures are per default gamma 2, than I get the right results, but I have seen, that I don't get the right reflection intensity. So, sRGB is a must.

[flipside]

Problem of cloud noises on surfaces

Yeah not only for reflections, it's very important for your GI too. With gamma 2.2 correction you will get much better lighting especially in interiors, where you probably needed all kinds of tricks to let the light travel far enough inside. With gamma 2.2, this is not a problem anymore, and also the oversaturated color bleed will look much better with gamma 2.2.

About subdivs, I usually use between 50 and 75. But I do mainly products, for interiors maybe there will be cases you need more. I always use the default sample value to get sharp GI details, but this needs high subdivs or you will get splotches.

You should do some test with the clr treshold and it's effect on IR map calculation time. You will see it is not linear at all, and values below 0.3 result in extremely higher times. I did that test in an older vray for max version, it's very possible the results don't apply anymore to current vray versions.

this was what it looked like. As you see, between 0.3 and 0.2 it is very tricky Of course the GI is better at 0.2, but I don't think the improvent justified the rendertime.




The 'samples' value can also affect rendertimes a lot. With high values, vray needs a lot of work to interpolate, so depending on the scene this can have a big impact.

But I see you use some kind of 'unviversal setting' method with your AA set to 1/50 so I guess rendertime is not always an issue for you. I like the maximum control so I use 1/4 and adjust all my material subdivs etc seperatly.

[Micha_cg]

Problem of cloud noises on surfaces

Thank you for the chart. Interesting.

I understand it so, at 0.3 and lower the color threshold start to work. At 0.3 it works a little bit, but with 0.2 it works at 80%. If we use 0.3, than the adaptive method is use only a little bit. But if I use 0.2, than I get what I want - quality in difficult areas. I learn from the chart, that between 0.3 and 0.2 I can choose, how adaptive the IM place the sample points. I would say 20% ... 80% adaptive. If I use values above 0.3, this adaptive process is disabled, other process are used more. I think, I will try values within this range in my next renderings.

I suppose so the adaptive mode of QMC sampler dosn't work for the IM subdivs or? So it could be, if you use a high subdiv value, every pixel in the image is sampled at this high count. Uninteresting areas of the image with smooth lighting will be sampled at a high count too.

[flipside]

Problem of cloud noises on surfaces

Don't know about the adaptive thing. But as far as I know this does affect ir map.

But the 3 thresholds in ir map simply control if an area needs additional sampling or not. It compares nearby samples and then uses the tresholds to decide if it needs more or not. The clr is the most important one as it compares colors of the samples, or in other words, differences in lighting. My graph is simplified a bit, after 0.5 it still goes down a bit.

It's not because you use high hsph subdivs, that each pixel will be sampled. This only controls the quality of each sample, it will not add samples if you use higher subdivs. Then i think the 'samples' value needs another name because it doesn't control the number of samples produced by vray at all! It should be called "number of interpolation samples". Try values of 8 or even lower, it will keep the GI detail much better, but it will therefore be less smooth, and so you'll need higher subdivs to clear that up etc... But if you're going after such detail, better use QMC GI

In vray for max we now have a combination of IR map and QMC GI, called 'detail enhancement'. This is a great thing, you can use low IR map settings, and then QMC GI will be used in detailed area's! Best addition to vray in years imo! Please bring this to vrayfor rhino

I would never go lower than 0.25, it's just not worth the time. Then you'd be better of using other min/max values, this does have a huge effect.

[Micha_cg]

Problem of cloud noises on surfaces

Thank you Wouter for the insights.

[Micha_cg]

Problem of cloud noises on surfaces

Upps, I can set the color threshold at 0.3 or 0.1 or 0 and the render times are not different. I see a little change at the samples only.

[flipside]

Problem of cloud noises on surfaces

Strange, it has a massive impact here:


clr=0.5, 14s


clr=0.2, 86s


You need a scene with fine detail to see the effect best, a groundplane with a cube on it will not show such a big difference of course

[Micha_cg]

Problem of cloud noises on surfaces

Thanks you Wouter. I see now - if I jump from 0.5 to 0.1, than I see the effect. It seems to be, that lower values than 0.1 dosn't make any change. I tested 0 and it looks like 0.1. For a more easy interface it would be enough to jump between five preset values 0.5,0.4 ... 0.1 or?

Other strange effect: scene: open box with white sky around. I find no parameter to get a smooth GI lighting. Subdivs 300 or 66, it's no difference. So, here the UI could be work with five fixed values too - 10-20-40-60-80 or?

Since I know, that light portals cause a wrong light distributation, I try to avoid this tool. If this could be fixed, than the light portal could be a must. (If you like, could you check the effect at 3DSMax please? I thought, I have seen the effect at a 3DSMax machine too some weeks befor.)

High settings, but still noisy


And multipass? If I set -3/-1 or -1, the difference is very small, only the render time of the multipass is longer. can we ignore the multipass?

Make it sense to use max rates above 1?

Did you every try to set all thresholds 0? I thought, I could set the IM in 100% adaptive IM sampling, but I it dosn't work.

So, it seems to be, that IM can not be set in a high qualiyt mode or? I thought, IM in high quality mode could be faster and cleaner than QMC. So, the IM is tool for low or meadium quality only or? If yes, than a limitation of the values could help to use it. Here a screenshot of Rhinoman that show a value limited UI.



So, I find no way to get good shadows and no artefacts. Here my scene file.

[flipside]

Problem of cloud noises on surfaces

Here micha, just found this on the vray forum:
http://www.cgarchitect.com/upclose/V...ngs_Part_I.pdf

If there is no difference between 66 or 300 subdivs, something is wrong, or you're not using a good testscene. Maybe the problem is that you use high amount of 'samples' value. Keep this at 20 while testing the subdivs. The samples value blurs your GI so if you want detail, you should LOWER this value. I would not limit the subdivs value to some presets, just because you test this value on the simplest scene. In a complex scene, going from 50 to 45 for example can help reduce rendertime a lot, without affecting the GI too much.

Fot the clr value, see the graph I showed earlier, you can see that below 0.1 has little effect on rendertime. you can also see that the best setting is 0.35, with best I mean best compromise. (vray is all about compromising)

About your light portals, these were created in the old days when nobody used LWF. So in these days, there was a problem that the skylight didn't penetrate an interior enough. So people used vraylights in windows to brighten up the scene. To make things easier, vlado implemented the portal mode, which camptured environment light, and bundles it and throws it inside. So this is what the portal does, get more light inside. But, 1 important setting is the 'ignore light normals', turn this OFF and you will see the difference will become a lot smaller. The ignore normals is a fake and I almost never use it, your light will shine as bright 90? to the right is it does straight ahead. Especially in your light portal test scene, this will have a huge effect.

It makes sense to use rates above 0 or 1. if you render a 800*600px scene with -3/0, it is the same as 400*300 with -2/1, or 200*150 with -1/2 and so on.

Multipass is not useless, because the second pass uses info from the first pass to calculate this second pass. The first pass is at a low resolution, so it can be calculated quickly, the GI is very rough, but the second pass will use this rough info and refine it. Then the third pass comes and uses the info from the second pass and so on. So in general, the passes should result in lower rendertimes. Of course this depends on the scene again, a simple scene will not benefit that much from it.

IM is not effective if you use all low tresholds. It is an approximation of the GI so don't try to get such clean and perfect results as QMC GI with high subdivs. It is indeed a tool for medium quality, but it depends on what you call medium quality. For many people, what you call medium is already very high. I would call IM high quality, and your quality needs I would call 'exact', like maxwell.

[flipside]

Problem of cloud noises on surfaces

In vray for max, there is an option called detail enhancement. This mixes the ir map with qmc GI so you get the best of both worlds. I think that is what you're after, and I guess (hope) it will be added in newer vray for rhino versions too?