Global Incremental rendering - it's would be wonderful

[Zlobnyi Serg]

Hi, all. Hi, vlado!
I have a great idea:
I think that's not a problem making many things rendering incrementally.
My proposals are to optionally select these things:
1) Make antialiasing renering incrementally. First, image renders without AA (and displays on the screen), second pass determines where more samples needed to be added to achieve good AA result (considering existsing algorythms and AA settings) and calculates it.
This decreases render-time consumption if we'll see some artifacts on early (first) pass.
2) The same we can do with IRmap. Now realized only incremental sampless adding to the irmap pass-by-pass. I think it's not a problem to make HSphere samples be incrementally increasing till the user see what result is good enougth.
All these thing can help us (VRay users) to do fast previews without changing render settings. If I'll see what image looks great I can give to render process more time.

How do you consider? Is this a useful abilities?
Vlado, please, tell what you think?

[vlado]

This is what VRay does to a large extent internally - for example, the adaptive AA first renders the buckets at low resoltuion, and then progressively adds samples where needed; the irradiance map also adapts the number of rays locally for each sample. However, there are limitations to applying this for the whole image. The main limitation is memory - incremental updates require additional memory, which can be prohibitive (that's why the image is rendered in buckets, and not in one pass, for example).

Best regards,
Vlado

[Zlobnyi Serg]

I understand these things because I'am a programmer and I clearly can build schemes of such incremental processes in my head
And I'am shure what memory consumption for this process is very small in comparison, for example, of amounts to store photon map.
Yes, for big resolutions, perhaps, this method will slowdown rendering, but in smaller (which are need for interiors, for example) - 1024x768 memory usage and time loss will pay to itself.

Very often I encounter such a situation: waiting for photons, irmap calculations of scene with many glosiness effects and after start of bucket rendering of final image I see some artifacts wich could be seen with dramatically lower settings of irmap, for example. Thus, incremental rendering with intervening image draft visualization can help users to reveal scene/modelling/settings errors in early stage.
Time saving for scene setup - this is the main aim of incremental process.

[Geddart]

Hey Zlobnyi Serg

Did you see the FPrime Video's? It's the same approach - refining the Quality step by step while keeping the user up to date all the time.
I'm the same opinion - it's very cool and elegant technique and it let's you judge the overall image very fast. I'm also pretty sure that there are ways to handle the memory consumption with this process.

BUT! I don't think it's the right time for the Chaos guys to implement that new render technique. Certainly large portions of Vray had to be rewritten. Remember, FPrime is designed from scratch for this adaptive rendering style - it won't be that easy for Vray. I think the Schedule of our Heroes is pretty full right now - first they have to manage 1.5 and maybe some bugfixes after it's release. (totally new features bring new bugs...) Then there is the Standalone Version and the Maya port which won't be easy at all. And finally there's a little Diamond called Aura waiting to be released - you see it's not the right time to rebuild Vray from the ground up.

But sure - for the future development of Vray this new technique really has to be considered. The more computational Speed we get, the more attractive this technique get's. Even now you can work with FPrime on small scenes almost in realtime - with full raytracing, GI, DOF and all the other nifty features. This really speed's up your daily rendering work and in a few month's this will work with larger scenes too. Maybe the Chaos Crew can think about it for Vray²... I just think for the 1.x cycle of Vray this is not an option.

regards

[tom schuelke]

Hi Sascha !

Nice to read from You while you sit here beside me at lichtecht. .

I am of Sascha Gedderts oppinion.

This adaptiv rendering process seems to be a great workflow enhancemant. the Videos at F-prime are realy impressive. maybe in future and maybe together with future Hardware acelleration this will improve renderers to the step , that we will be able to play with light, colours and effects as we never could before. interactiv scene and light testing

so if this would be a future feature of Vray, it would be a big step in the right direction.

So maybe Vlado, this could be an option for smal scenes ?

greetings

Tom

[Zlobnyi Serg]

2Geddart: The directions of VRay improvements you mentioned is ofcourse important, but inremental features needs less efforts than some planned jobs, such as standalone version and maya port.
So, chaosgroup can give interesting and attractive feature to this render without extreme efforts.
For example, as programmer I can consider what implementation of incremental antialiasing (I mentioned it in my first post) can take from 1 to 7 days of one programmer work... This isn't big time in comparison of attraction of such feature...