Over rendering

[Rich Rosemann]

Generally, I read that to solve a rendering problem, render longer to achive an acceptable result.

 

My question is, can an image be over rendered waiting for individual materials to complete?

Is there a point of "Rendering Over-saturation" for individual materials (my description)?

Does a material reach a point of maxium rendering and simply stop contributing to the image?

Is this point determined by the image resolution, or the material sizes (in pixals?)?

 

Playing with fog.  Sooo cool.

 

Rich Rosemann

 

 

 

 

 

[Roy Hirshkowitz]

You can't really hurt anything by letting it continue.  You can certainly get to the point where you stop being able to tell the difference between your current "estimate", the next "estimate", and even a completely converged "solution."  Convergence depends on many factors, although, for most scenes, its more dependent on lighting conditions and material properties than on texture or image resolution.

[Rich Rosemann]

Thanks for the reply, Roy. 

 

Rich Rosemann

 

 

[George Ioannidis]

My question (more rhetorically meant Roy)  on that:

Would it have any point if nXt could determine areas of particular rendering, which are more difficult to solve and need a lot of passes to be cleared from noise so it could stay focused mostly on them rather than calculate over and over all image pixels, as changes on "difficult" areas barely affect the "easy" ones - I thing the opposite is happening - they stay almost unchanged from the beginning of the rendering process anyway? Could something like this speed up the rendering somehow?

This applies mostly to exterior renderings from my experience.

[Ros. Riha]

this is good idea... maybe you could mimic it now via "render region" - IOW: you run the whole scene for 50passes, then you start "render region" of the part, which proved difficult to resolve - despite the fact that you will be rendering that part twice, you might end-up with an overall gain...

[Marc Chaumier]

Good idea, the regions where there is no direct light would ask for more passes. For example,10 passes on these, when 1 pass is enough for regions under direct light. And so on blurry reflections, transparency... But it need the engine to evaluate the difficulty of the work and do a ratio between the difficulties to decide the number of passes. And the overall speed of render could, probably, be decreased, to have the same average of quality, averywhere in the picture.

[akinlolu olugboji]

How would you advise me apply this method, to a work like like this, already cooking for 1227 passes,

[Roy Hirshkowitz]

Don't do it.

[Ros. Riha]

In this case it seems to be rather the question of materials to me - I do not see why this scene needs this many passes

[Roy Hirshkowitz]

Difficult lighting.  One small opening, much of which is obscured.

[Chas / Aug Posekany]

Seems to me the rendering will eventually resolve to your liking.

The bump on the tile backsplash with the higher reflection and the interior daylight scenerio is just very difficult to resolve.

IMHO

[Roy Hirshkowitz]

It's a great question and an area of ongoing research.  In the field of Monte Carlo integration it's known as adaptive sampling.  It's not an easy thing to implement.  Most of the simple ideas such as "work on the dark areas more" don't tend to work very well.  One of the more promising recent approaches is using a measure called Tsallis entropy to decide which pixels need more work.