www.kraytracing.com

Hi! I'm not sure where to post this on the forum, or even if this is allowed, because it is about another render engine. If this is against the rules please let me know and I'll remove the post.

I just wanted to post some render I did with Octane this week. I hope I will be doing all of this work with Kray 3 soon. At the moment the glass shader and lack of subpixel displacement make this process a lot slower in LW + Kray. The speed of setup for these renders is amazing, and I say that as someone who's only using Octane for a day or 3.

It really is almost:
1) Create Bottle Model and ground plane.
2) Apply Glass shader. Apply "print" shader - these just work practically "out of the box"
3) Make mapping to say which part is glass and which part is printed.
4) Make and apply displacement mapping.
5) Render.

On the bad side: noise. Still, the speed of process makes up for a lot. (edit: the renders are the original renders, not rescaled, so you can judge for yourself about the noise).
(edit 2: these renders are almost straight out of Octane. The first two have dissolve noise added in Photoshop to get a metallic carpaint sparkle, all others just have some contrast adjustment done in Photoshop).

I forgot one more "bad side". Almost everything (except for some procedural noise shaders) needs a UV map in place. For this type of model that's not really a problem of course, since a simple cylindrical projection takes care of almost everything.

Thanks for renders.
This glass fission of colors G. told, we we have in K3 Core.
Yes, the noise is annoying not because of noise...is annoying because I know when I will put normal interior scene with complex light and objects it will take forever whole scene render. Studio renders in Path Tracing always render fast because is nothing to calculate there. Length of render drastically grows with amount of bounce, amount of lights, amount of objects and amount of reflection blur. Rendering interior must be done with minimum 7-10 bounce light and each bounce light for Path Tracing is killer for time render. Trying render interior with recurse under 5 affect realism.

You are very right. I'm putting a faster GPU in my PC tomorrow because I will be using Octane more in the future, but only for studio setups and bottle packshots like this. Even after one week of working with Octane it's obvious to me that interior shots with Octane are a world of pain. Besides noise and lighting quality vs. speed it's just too cumbersome to create UV-maps for everything (no "automatic sizing" in Octane!) and to need to wait for each test render for the image to be copied from main RAM to GPU RAM.

can i have the last bottle one pls ? :3 .

lightwave3d@hotmail.fr thank u sir. i want to try it

Alright! I'll put a WeTransfer link up here tomorrow. I was thinking about doing some Kray tests myself, but the more people work on glass in Kray the better. I can't upload it now because the files are at the studio and I'm at home, but tomorrow I'll do it!

Here's the link guys - the more people trying to get a good glass look in Kray the merrier, so download away and post your tests here!

http://we.tl/kQ4CEkguzo

Cheers, Thomas

I tweaked your render settings for you....
totally noiseless image with only 75% of the samples you used and using the faster kernel...should render in half the time

Coherent ratio...GI clamp and caustic blur are the keys here

Janusz Biela wrote: Rendering interior must be done with minimum 7-10 bounce light and each bounce light for Path Tracing is killer for time render. Trying render interior with recurse under 5 affect realism.

The bounce for PT and the bounce for photon mapping are 2 totally different things.

The bounce for photon mapping is an interpolated rough approximation GI of the photons.... followed by a single bounce final gather pass for fine details.

Path tracing is a non interpolated every pixel bounce. In PT every sample takes a different path...where in photon mapping, a single photon takes only 1 path. So a PT pixel is a culmination of multiple bounces, and multiple paths.

And a photon mapping pixel is a culmination of 1 photon, with multiple bounces, and 1 bounce final gather
Even though more bounces is better in interior...you get more realistic results with less bounces with PT. The strategy leads to more realism....but more noise

In my previous render, I had to LOWER my settings to 2 bounces to simulate your render of 5 bounces+

Completely different beasts.....definitely agree with you that photon mapping is a better strategy for interiors...and PT can be very noisy...but you cannot equate the number of bounces between the two renders.
5 is a pretty good interior starting point however

Alright, thanks MdHarrington, for that scene! No computer access here but will test tomorrow!

Idea of Bounce light should be the same in Path Tracing and Photon Map - otherwise is not according to physics. Of course there are many differences but formula is exactly the same.
Coders tried avoid real bounce lights by special tricks (multiply, limited bounce lights, Occlusion with lob bounce light and other) but this always brings artificiality in renders more or less visible.
Big difference is with amount of data to calculate. In Path Tracing you have no interpolation and this mean rendering forever - rays per pixel. Each ray is calculate separately, more light - more calculations, more bounce light - more calculations, more corners - more calculations.

In Photon Map reduction of data is done by cells (GI resolution) - each cell created is from point which represents position of photon. Photos comes in limited amounts from Camera (Lightmap) or more accurate but longer from lights sources (Photonmap).
So instead to calculate billions of photons in Photon Map you calculate limited amount: 10.000 100.000 etc. Also GI Sampling in Photon Map (Final Gathering) is interpolated so instead calculate billions of samples Final Gathering calculate thousands GI samples and missing part comes from interpolation between samples.

Exist also hybrid mode (This is the best from examples above) QMC mode where: photons position/information comes from interpolated data (Photon Map) but sampling is calculated with very similar method like Path Tracing (of course this is not MTL, but Quasi Monte Carlo is really good for rendering). This brings of course noise similar to Path Tracing because also there is rays per pixel.

Important thing is that when you use open area in render is enough to use 2-3 bounce lights in Path Tracing (also in basic studio renders) This brings false feeling is Path Tracing fast and no noisy. Most of the photons in that situations will bounce 1-2 times and end-up in "space" leaving information about GI. Situation DRASTICALLY change in closed area (Interiors) when same photon must bounce a couple of times and most of them must be calculated fully.

Problem with Photon Map is accuracy. Interpolation make a lot problems. Why to correct those errors in Kray exist special Path Tracing engine which shoot EXTRA PT photons in corners. Of course with this tool render is longer but more accurate.

It will be never situation when Photon Mapping will have similar quality like Path Tracing...but we never know

Janusz Biela wrote:Idea of Bounce light should be the same in Path Tracing and Photon Map - otherwise is not according to physics.

not arguing the laws of physics....just that the bounces for 1 are interpolated and using 1 path...and PT using multiple paths (1 per sample). Nothing to do with physics.
The most important bounces IMHO are the last 3....as even with a 50% reflective surface....it's contribution to GI is down to 1.5% at the 5th bounce, and PT will acquire more surface samples for these last 3 as they are non interpolated. So the trade off in lighting contribution would be countered by the detail of the most important bounces.
The hybrid QMC mode is indeed interesting...best of both worlds.

There is talk of some GPU accelerated photon mapping coming soon....I think hybrid renders are the only way to go, as pure brute force path tracing is only practical in certain areas (outdoor and massive polygons are best use for them). The photon mapping GPU algorithms are already out there....but the brute force purists are hung up on 'unbiased' renders.

Looking forward to see what K3 has to offer

I doubt about photon mapping done by GPU...the code is too big for CUDA cores. fragmentation and after that de-fragmentation will brings only problems . I am not coder but what I know, the Path Tracing code is very simple that's why is easy calculate by GPU in hundreds/thousands cores in parallel . I think future is only in hybrids CPU/GPU. Some effects are faster calculated by CPU some by GPU.
G. told me is difficult to split job between GPU/CPU in the same time but for good coder still possible. K3 code has GPU rendering (not finished yet and not activated) but I can test only basic things by GPU. Of course target is hybrid.

there have been a handful of GPU photon mappers already written...here is one example
https://vimeo.com/81835028
https://github.com/ishaan13/PhotonMapper

Nowhere near as complex as Kray code for sure....but basic photon mapping accelerated by the GPU with the K/D tree calculated by the CPU

some Octane devs said they would be looking into this or other implementations...possibly Gradient domain path tracing
(look on pg 11 of the PDF)
https://mediatech.aalto.fi/publications ... _paper.pdf

Like you see from links this is progressive PM (perhaps no enough memory for execute code in CUDA cores) so only small portion of data can be calculated. In the end this method can be faster then Path Tracing (of course) but slower (hard to day how much...can be unprofitable) then traditional Photon Mapping. In my opinion no future for this. But I am not coder so technically I can be wrong.