Caustics quality

[Thomas An._cg]

Hi,

This images uses Irradiance map + Light Cache(single frame)

What parameters should I concentrate on to improve the smoothness of the caustic patterns ? Is it the min/max rate ?
Should this image be done in about 20min ? (is 2hours a normal range when caustics are involved ?)

[vlado]

Caustics quality

Did you get the caustics through GI alone, or you enabled them from the Caustics options separately? Depending on which method you chose, the settings that you need to adjust are different.

As for the render time, it also depends on what other settings you have chosen (e.g. antialiasing engine etc). For photon-mapped caustics, 2h is definitely too long.

Best regards,
Vlado

[Thomas An._cg]

Caustics quality

Thanks for the reply

The caustics are from GI alone (the indirect illumination tab). There are two area lights in the scene with the option to "store with irradiance map". The image sampler used was adaptive QMC.

For photon-mapped caustics, 2h is definitely too long

I am not yet sure how Vray works, but shouldn't anything relating to photons be handled by the photon map ? However in this case there was only Irradiance+LightCache enabled ... so I am not sure if the caustics are "photon-mapped". Are they ?

[flipside]

Caustics quality

The caustics are photon mapped if you enabled caustics in the caustics rollout. Photon mapped caustics are a way of speeding up difficult caustic patterns.

You used only GI caustics, so the settings to concentrate on are mainly the IR map settings. Especially clr treshold will be important (of course also min/max and hsph, but these should not be set too high, like -1/2 or something, a normal -3/0 with low clr will create already a lot of detail in your GI lighting and thus also caustics)

I'm interested in rendering that model, to tweak the settings to the max!

But i see you use glossy refl on the groundplane, and dof too, this will add to the rendertime also of course.

[Thomas An._cg]

Caustics quality

Thank you Wouter,

I was beginning to suspect the color setting, but was having other problems (because of my newbie-ness :P ). Also, all ground surfaces are matte... there are no reflective surfaces (other than the metal rings and coins)

I will be glad to send you the scene. Can you PM me with your email ?

Thanks
Thomas

EDIT: File Sent

[flipside]

Caustics quality

I'm doing some tests with IR map + LC, QMC GI only, PPT and photon mapped caustics.

Can I see your maxwell result somewhere, with rendertimes so I can compare a bit? I'm interested in that result, how well maxwell handles this and in what timeframe.

[Thomas An._cg]

Caustics quality

Sure, but I would like to hold off, before showing you the Maxwell image, because I don't want it to be used as a reference to tweak the Vray image. The idea is that it should be possible to do this in Vray without a photo reference or a Maxwell image.

It is a good thing that Vray now has PPT, because we can use the PPT to get a preliminary idea of how strong should be the caustics and the secondary illumination (without resorting to photos). Let me tell you, the PPT test I did (5hour test, but still needed more) was getting close to the Maxwell quality of illumination.

Now the Maxwell image has slightly different emitter intensities (I have changed them slightly ... actually, in reality, I didn't know how many watts or lumens are the units of the Vray light so I was adjusting them arbitrarily).

In any case the Maxwell result (which I will show you later) took 63hours for it to look smooth enough (raw output without a noise reducer) ... now if we can achieve the same realism with Vray in say 40min and without the use of a photo reference then I will be blown away !

[flipside]

Caustics quality

Ok it's a challenge

But I changed some stuff in the file, also on lighting... I guess I need to re-add dof to make it fair to maxwell.

One big difference which has been pointed out in the vray forum, is that maxwell by default gamma corrects the images, which can be done in vray by using gamma color mapping. I think I will do so too, because I think in your test with similar lights as in maxwell, the vray image looks much darker no?

Never mind, I'm just gonna make a nice image with vray, according to what I think is nice (and correct)

[flipside]

Caustics quality

Here are two tests, all on dual Xeon 2.8Ghz, 3GB Ram

1. ir map + LC. All caustics are GI caustics: 57m11s



2. ir map + LC + photon mapped caustics: 54m48s



I didn't tweak settings to find an optimal result, I just did some tests and turned up the settings a lot.

The first one with ir map is difficult, because IR map is not the best enging to capture really sharp and crisp detail. I lowered the (interp) samples value so that the sample values are not blurred out that much. Maybe I should have used another interploation type like delone triangulation, which is not a blurry method as described in the vray manual.

The second one uses photon mapped caustics. So the caustics are not calculated by the ir map. This has the advantage that you can define caustics quality sperate from IR map. So I used lower ir map settings here, but there is some time needed to compute the caustics, resulting in similar overall rendertime. The caustics are better though in the same timeframe. I am not very familiar with caustics settings, so probably also here rendertime can be cut down. Especially on IR map I could have used lower settings for the same quality I think.

[Thomas An._cg]

Caustics quality



Wouter, Thank You for giving this a try !
I like your second image.

The caustics are a bit of a challenge. I have been doing some tests over the weekend as well with irradiance map and light cache, but it seems the only way is to turn up the settings quite high.

In your second image... yes I see the cautics are improved quite a bit. So you used the caustics rollout tab ? and you unticked the Reflect/refract caustics from the "Indirect Illumination" tab ?

I have been studing this page quite intently:
http://www.spot3d.com/vray/help/VRay...xamples_GI.htm
... and the IrrMap+LightCache are fast but did not give the most accurate result (judging by the shadow distribution under the chrome sphere). It seems that IrrMap+PhotnMap as well as PPT had the best shadow distribution (QMC would too, if it had more bounces). However, my photon map tests were atrocious, so I gave up on IrrMap+PhotonMap+retrace corners.

Well, here is the Maxwell image. The bokeh was off due to a bug in the polygonal thingy, but the BSDF defined materials are hard to beat. The Vray caustics are quite close (and hopefully one day I should be able to make them milky smooth in much less time than 63 hours )

[flipside]

Caustics quality

Yeah, ir map is not the best to compute this kind of caustics. I did a test with delone triangulation for interploation type, an it was slightly better in the coin caustics, but worse in the caustics coming from the ring on the outside.

By using QMC GI for first and secondary bounces, I get quite decent results, but long rendertimes (not 63H though!). I had a 1h50min test which was sharp, but noisy (but sharp noise!). I think with double or triple rendertimes it would be very good. QMC for first and LC for second could be a good option I think. But then I don't want to think about larger resolutions...

I'm doing ppt right now, rendering for 2h but still far from good, the caustics are coming trough in sharp white pixels, still needs lots of work.

Don't get me wrong, I never used photon map for GI!!! I only used photons for caustics. So in that test, I used IR map for first, LC for secondary, and then caustics turned on in that caustics tab. Then in the lights options, you need to alter the 'caustics subdivs' (not the photon subdivs!!!)

It doesn't matter if you turn off GI caustics or not, it seems that vray does that for you if you turn on caustics in the caustics rollout.

I used these caustics settings:
max photons=60
max density=0.01
multiplier=1
search distance=0.75

The last one is important, because when you use higher distance, it will blur more surrounding photons together.
But you need to use high caustic subdivs on your lights, I used about 5000 to 7500!! This is hard to do test renders, because when you set it to 750 or 1000, you don't know at all how the caustics will look like. So you need to test it at 5000 and see if it's enough...

So to summarize, if you don't want to test too much with settings, I would suggest using GMC GI with 50 to 100 subdivs and you'll get a render of 2 to 8 hours.

If you want great caustics, I would use photon mapped caustics (caustics rollout!) and ir map in first with LC in secondary bounces. You can also render in larger resolutions without the caustics creation taking longer, same for IR map if you lower the min/max by 1 each time you double the resolution.

In 99% of the cases, reflection caustics are unwanted, I usually turn if of in the GI rollout and never use photon mapped caustics. Only in rare cases like this image you really want nice caustics. I have never needed to render high quality caustics for any of my clients

[vlado]

Caustics quality

Thomas, would it be possible to get this scene so that I can play with it?

Best regards,
Vlado

[flipside]

Caustics quality

Here's my last test.
Ir map for first bounce
LC for secondary
Photon mapped caustics: bumped the caustic subdivs of both lights to 15000

Rendertime about 1h30minutes. It won't get much cleaner I think, but wait untill Vlado trashes all our rendertimes with a noise free 10minutes render

[Thomas An._cg]

Caustics quality

Wouter,

woo This last one is a very clean render and the time is not bad at all (actually 1.5hours is awesome, I am very happy with your result).

The only thing that bothers me is how come Vray does not reflect the intense glow of the righthand ring onto the righthand coin). In my tests I have seen that it tries to do so, it actually tryes to show the reflection but it is being filtered out by the QMC sampler (we have to set the noise threshold really really low to start seeing something)

[flipside]

Caustics quality

Well I know the answer... Under color mapping, there is an option called 'clamp output'. This clamps all color values to something in the 0-255 range. I have this turned on.

When you turn it off, it renders correctly, but there is a huge drawback. Because of the strong lighting values (especially here with the 125 intensity of the small rect light) the reflections of this light really looks bad (jagged).

Here's one of my early tests showing the problem. If I keep the clamp output turned off, I can't get nice AA on all strong reflection regions.