Animation rendering homogeneity

[Radiance]

I would be interested in some actual data/examples of the rng seed being set the same vs. different between frames being better or worse from one another.

The scene is changing regardless, shouldn't the result be roughly the same (as in, there is no consequence to using different seeds?)

Hey,
The noise will be different, eg it has a different seed, so that grain positions/locations will be different.
The level of variance will be identical though, so it won't help in any way.

Radiance

[Radiance]

Here is an example of some pictures in my animation. As you can see, the result is really too various to give a comfortable result!
(it's not obvious like this : to be aware of it, download and look at them one after the other)

Hey,

Yeah, just be patient for the next weekly build, i've got heaps of stuff to do currently

Radiance

[ackbahr]

Yeah, just be patient for the next weekly build, i've got heaps of stuff to do currently

Oh, I'm very sorry if you had the feeling I was trying to put even more pressure on you; I'm sure everyone suggest e.g. expects his little detail ASAP, and that you've got more than enough to do! I had just realised I had not answered zsouthboy's message asking for examples....
Anyway thanks again for all the work on luxrender and here answering our messages, must be 26h days! Looking forward for exciting new stuff in weekly builds to come....
Ack

[ackbahr]

For those of you who might be interested, here the animation I'm trying to render : I give you the complete one as rendered by Blender's internal engine, and the little pieces rendered with Luxrender (sorry, to monitor results I set a random order for frame rendering.
Enjoy (if it's worth anything at all....)!
Ack

[Ratow]

Indeed, i've been thinking about adding a command line flag/switch to ensure same seeds for all threads for rendering animations, i'll see if i can add one asap and maybe it will be in the next weekly build.

Ok, a specific flag for using the same seed is now available (-f or --fixedseed). However, one still should use only one thread in this case since the thread scheduling is not deterministic. I wonder if we should put a warning if more than one thread is used with the -f flag or if the effects aren't that noticeable.

-Ratow

[Radiance]

Hey,

As long as you're giving the same seeds every time, and the user is not adding/removing any threads while rendering things should be identical all the time not ?

Radiance

[Lord Crc]

Well, due to scheduling by the OS, one thread might do 70% of the work while the other just does 30%. Then there's a difference in how much was rendered with seed 1 vs seed 2. However how large this effect is in real life I don't know.

[ackbahr]

Great to have so quick answers to suggestions and demands! But I don't really understand : when Radiance first told us about this same seed thing, I thought it was something happening during export (from Blender for me), so all pictures would react the same. But here I understand it's something happening when starting luxconsole? But if I render sequentially, how is one luxconsole instance supposed to know what the previous's random seed was?
Wouldn't it be possible to have a non random, homogenuous "map" of what pixels should be rendered, that one could invoke even from several computers?

[Ratow]

Wouldn't it be possible to have a non random, homogenuous "map" of what pixels should be rendered, that one could invoke even from several computers?

This is more or less what is achieved with the -f flag. Lux will use a fixed seed for its pseudo-random generator and so the same pixels (and light paths) will be rendered anytime, anywhere.

-Ratow

[ackbahr]

Ah, ok, so this *should* work even if I render from multiple computers? Well, as for my dual core machine, maybe I can run two one-thread renders at the same time....
Thanks again, it's great!