Is there any further development on Irradiance map to make it more intelligent?

[Lupaz]

I understand that many people use brute force.
Irradiance map GI is still a faster approach as far as I can tell.

The only "problem" I see is that setting subdivisions and interpolation samples to get a nice balance of accuracy, depending on the parameters Low, mid, high, is still too manually done.

Is there a chance of getting the software to do this balance? How about also taking into consideration the type of scene?

Thanks.

[3LP]

Vray quick settings?

Stan

[vlado]

The only "problem" I see is that setting subdivisions and interpolation samples to get a nice balance of accuracy, depending on the parameters Low, mid, high, is still too manually done.

The next service pack includes a new adaptive algorithm for the DMC sampler, which reduces the need to adjust the subdivisions. Interpolation samples I usually leave at the default, I haven't found too many good reasons to change it.

What types of adjustments do you usually do?

Best regards,
Vlado

[Lupaz]

The reason why I adjust interpolation samples is because, as far as I know, the smaller the number the more accurate the shadows. This gives a nice contrast to the image that you would get with brute force.

However, it did happen to me that I went too low and ended up with a noisy image. I always render regions to make sure everything is ok. But some times there are some tricky areas hard to predict.

At the same time, if I want to reduce the number in interpolation samples, I need to raise the subdivisions to compensate, especially if I choose a "high" preset. There is no method, again, as far as I know, to have the "correct" balance.

Also, more subdivisions are needed for closed spaces.

I was wondering if there were plans on doing the amazing job you guys did with DMC and AA, for cached maps.

Regarding light cache, I heard around the foum that a good number for subdivd is usually the amount of pixels the image has on the largest side. If this is true, I wonder why it's not done automatically by vray.

Thanks,

[vlado]

The reason why I adjust interpolation samples is because, as far as I know, the smaller the number the more accurate the shadows. This gives a nice contrast to the image that you would get with brute force.

Hm, I can see why you could want to increase the interpolation samples to get smoother GI, you are the first user that is decreasing them How many are you using (the default is 20)?

Also, more subdivisions are needed for closed spaces.

Let's hope the new adaptive method will help a bit with that.

I was wondering if there were plans on doing the amazing job you guys did with DMC and AA, for cached maps.

I would love to, but with things gradually moving away towards more brute force approaches, it looked like optimizing that use case is more important than spending time on the irradiance map.

Regarding light cache, I heard around the foum that a good number for subdivd is usually the amount of pixels the image has on the largest side. If this is true, I wonder why it's not done automatically by vray.

Because it's a bad rule There's no reason whatsoever to change the light cache subdivs depending on the resolution. I'm totally mystified how that myth came into existence.

Best regards,
Vlado

[Lupaz]

Hm, I can see why you could want to increase the interpolation samples to get smoother GI

Isn't smoother GI = blurrier GI when interpolation samples are higher?

How many are you using (the default is 20)?

I'm trying to push it lower than that to get better, more detailed, shadows.

Because it's a bad rule There's no reason whatsoever to change the light cache subdivs depending on the resolution. I'm totally mystified how that myth came into existence.

Is there a rule at all? To be honest I don't notice the difference in quality between 300 subdivs or 3000...

[Alex_M]

You mean brute force for primary bounces? Really? I've found that it's still very slow to use BF for primary bounces (in case I'm missing something that optimizes things). I do know that quite a few people now use BF for secondary bounces instead of light cache but for primary? I imagine that the percentage of cases where you need that much accuracy that you'll use BF for primary is really really small or?

I understand that many people use brute force.
Irradiance map GI is still a faster approach as far as I can tell.
......

[vlado]

Isn't smoother GI = blurrier GI when interpolation samples are higher?

Yes. Many people do that to get fast but smooth renders.

I'm trying to push it lower than that to get better, more detailed, shadows.

How much lower do you go? 10? 5?

Is there a rule at all? To be honest I don't notice the difference in quality between 300 subdivs or 3000...

Use 1000 for still images and 3000 for animations. (You can go lower for fast previews, but it may cause blotches near walls and flickering in animation).

Best regards,
Vlado

[Lupaz]

Yes, I meant for primary bounces.
Just as you, I think it's much slower than irradiance. I'm trying to stay away form BF as much as possible. That's way my questions.

[Lupaz]

Yes. Many people do that to get fast but smooth renders.

How much lower do you go? 10? 5?

I'd go as low as I can. If I can get sharper results (more accurate shadows) by lowering the number, why not trying the lower I can get without getting splotches?

[super gnu]

i also reduce interpolation samples wherever possible, as mentioned, you get a sharper result at the price of increased blotchiness.

i dint think its such an unusual thing to do. its not like people dont want to get a BF like result with imap.

i dont think ive ever managed to go below 10 on any production scene though, the necessary imap subdivs get prohibitively high.


i also agree however that things are moving away from such heavily biased methods, simply due to less setup time involved in using BF.

having said that, for flythrough animations, imap is still so far ahead of BF in speed that its embarrasing.. its the one time i would still *always* use imap.

one feature that ive often thought would be nice:

Cached BF.

that way you could have the sharpness of full BF, but the speed advantages of multiframe incremental for flythroughs. i imagine the caches would be pretty huge though?

alternatively a set of settings in the imap that would be equivalent to full BF in speed and quality.. - ive tried using 0,0 for min/max and 1 interp.samples, but its so much slower than BF that its not really practical, and despite the low min max, still appears to have blotches larger than a pixel.


edit: can any of the speedups that have been applied to the BF engine be applied to the imap engine? i guess there is a lot of overlap in the code? or not? i assume its not been updated since as mentioned, with settings of 0,0 its *way* slower than BF.