Just wondering about KeyShot tech under the hood...

[DriesV]

I was wondering...and I know it's not something that is typically discussed/advertised by Luxion...

• What technologies set KeyShot apart from competing products?
• What kind of rendering algorithms are under the hood?
• How will these technologies enable KeyShot to keep pushing the envelope in the rendering biz?

Is this information that Luxion can share (in any concise manner)?
I'm just interested in this sort of stuff...

Dries

[Despot]

Hi Dries,

Well, I don't think KeyShot uses path tracing / Metropolis Light Transport like Maxwell / Fryrender does... which means it's not unbiased...

I do know it uses photon mapping for lighting calculations, look at this Wikipedia link, recognise a name ? 

http://en.wikipedia.org/wiki/Photon_mapping

J

[DriesV]

Hi John,

I know about the photon mapping bit (it's also mentioned in the KS tech specs).
I think the whole 'unbiased' thing is not saying at all if the renderer is achieving a physically accurate (cfr. correct?) solution or not. 'Unbiased' is not necessarily more correct than 'biased'. Correct me if I'm wrong...
I think these are far too bloated terms.

In my experience with both 'biased' and 'unbiased' (oh, I do hate the terms ...) render engines is that in the latter the amount of error is somewhat more predictable. Give it 1 minute and it will look grainy. Give it half a day and it might start to look okay.
I do suspect KeyShot uses some sort of progressive photon mapping for the real time rendering, which in theory can be just as physically correct as most of the 'unbiased' (*aargh*) render engines.
The more photons you throw around, the more correct the rendering will be.

Dries

[Despot]

I think unbiased means that the renderer will always converge to the correct solution... meaning there are no errors in the radiance calculation... meaning physically accurate output

But whether you use a biased or unbiased render engine it doesn't mean to say that the renders produced are correct visually...

I've done work in MW that looks a bit iffy, but the same scene in KS looks amazing, and the other way around

J

[DriesV]

I think -in the case of KeyShot- the only major error that can be introduced is due to the size of the photon map (amount of photons).
But since (I think) KeyShot is using progressive photon mapping, it really behaves like an 'unbiased' renderer in realtime mode. Also converging to a correct solutions as the number of photons is progressively increased.

Dries

[PhilippeV8]

What the heck are you guys discussing ... I have no clue 

[Despot]

Also converging to a correct solutions as the number of photons is progressively increased

No, because it is progressive, it doesnt then follow that it is going to converge to the  correct solution everytime - I use Cinema 4D R14 and from version R13 I think, they introduced a 'physical' renderer with a 'progressive' option... but it is still a biased render engine regardless...

Only unbiased renderers produce physically/mathematically correct output , KS is biased...

Think we need Henrik as a referee... 

J

[Despot]

In fact Henriks probably watching this thread laughing his a** off at us mere mortals discussing his area of expertise like we know what we're talking about...

J

[KeyShot]

KeyShot uses a hybrid approach. It is not a algorithm that you can find in papers or text books We do have several new things in the pipeline, so you can expect better performance in some of the more tricky lighting setups. Currently, KeyShot is highly optimized for product shots.

Regarding unbiased vs. biased / consistent then there has been a general misunderstanding that unbiased is correct - this is not true. Unbiased means that on average the rendered image is correct - it does not guarantee convergence to a correct image. Consistent and biased algorithms on the other hand will converge to the correct result, but they may have bias (systematic errors such as blurry lighting details) that go away as more compute time is used.

In general consistent and biased algorithms will converge faster than unbiased algorithms, but they can be harder to control. An unbiased algorithm such as path tracing converge quite slowly, but is easy to implement and has been favored by many GPU implementations.

Also, physically correct is a term frequently used by many rendering engines, but KeyShot still features the only lighting simulation engine that has been verified independently by CIE (International Commission on Illumination) in report CIE 171:2006.

[DriesV]

Great!
Thanks for the insight, Henrik.

Dries

[DriesV]

Here's a recent and rather interesting read about bias in rendering.
http://cs.au.dk/~toshiya/misc.pdf

The paper does a very good job of explaining why unbiased rendering isn't the holy grail of computer graphics.
So for once and for all:
*Unbiased methods are NOT more accurate than biased ones.
*Unbiased rendering does NOT always converge to the correct solution.

note: you don't need to understand the mathematics to grasp the implications of the arguments.

Dries

[Speedster]

Well, I don't understand any of this.  But I do know I'm "biased" for KeyShot!
Bill G

[DriesV]

Well, I don't understand any of this.  But I do know I'm "biased" for KeyShot!
Bill G

Well, I understand enough of it to realize that a lot of the (internet) buzz surrounding unbiased rendering -and its assumed superiority- is at least ungrounded.

Dries