>> I found under some conditions you get a visible pattern to the noise
>> when doing a low quality "preview" render with no_cache. It's almost as
>> if the random number generator is seeded off the pixel coordinate or
>> something? Is this possible or am I seeing patterns where none exist?
>>
>
> A simple way to check that: Launch a 2 or 3 frames animation with no
> changes between the frames.
> If there is a repeatable pattern, the images will show the same pattern.
> If the pattern is just random noise, it will change from frame to frame.

OK it's the same every frame, I'll post an image in p.b.i.

Post a reply to this message

>>> I found under some conditions you get a visible pattern to the noise
>>> when doing a low quality "preview" render with no_cache. It's almost as
>>> if the random number generator is seeded off the pixel coordinate or
>>> something? Is this possible or am I seeing patterns where none exist?
>>>
>>
>> A simple way to check that: Launch a 2 or 3 frames animation with no
>> changes between the frames.
>> If there is a repeatable pattern, the images will show the same pattern.
>> If the pattern is just random noise, it will change from frame to frame.
>
> OK it's the same every frame, I'll post an image in p.b.i.

Actually I just realised this has an implication, that you can't render 
N times on one machine (or I assume on N machines) and then average the 
result. Should there be a way to set the RNG seed that is used for 
choosing the direction to shoot rays?

Post a reply to this message

Am 27.08.2014 08:57, schrieb scott:
>>>> I found under some conditions you get a visible pattern to the noise
>>>> when doing a low quality "preview" render with no_cache. It's almost as
>>>> if the random number generator is seeded off the pixel coordinate or
>>>> something? Is this possible or am I seeing patterns where none exist?
>>>>
>>>
>>> A simple way to check that: Launch a 2 or 3 frames animation with no
>>> changes between the frames.
>>> If there is a repeatable pattern, the images will show the same pattern.
>>> If the pattern is just random noise, it will change from frame to frame.
>>
>> OK it's the same every frame, I'll post an image in p.b.i.
>
> Actually I just realised this has an implication, that you can't render
> N times on one machine (or I assume on N machines) and then average the
> result. Should there be a way to set the RNG seed that is used for
> choosing the direction to shoot rays?

By default UberPOV will seed each thread's stochastic RNG with the 
current time plus thread number, but I guess I should change this so 
that a render started 1 second later doesn't give you the same seeds for 
threads 1..N as the previous one for threads 0..(N-1), and I should also 
look into why this doesn't seem to work for animations.

That aside, you can always specify a custom seed (or, more precisely, 
seed base, as UberPOV will still add the thread number to it) using the 
"+SSn" or "Stochastic_Seed=n" option.

Post a reply to this message

> By default UberPOV will seed each thread's stochastic RNG with the
> current time plus thread number,

Hmmm, something strange is going on, when I look at the difference 
between the exact same scene rendered a few minutes apart there are 
about 10% of blocks that are totally identical, the rest are random 
noise. See image I just posted to p.b.i.

> That aside, you can always specify a custom seed (or, more precisely,
> seed base, as UberPOV will still add the thread number to it) using the
> "+SSn" or "Stochastic_Seed=n" option.

It doesn't recognise the +SS option here (beta 7 release).

Post a reply to this message

On 28/08/2014 10:12, scott wrote:
>> By default UberPOV will seed each thread's stochastic RNG with the
>> current time plus thread number,
>
> Hmmm, something strange is going on, when I look at the difference
> between the exact same scene rendered a few minutes apart there are
> about 10% of blocks that are totally identical, the rest are random
> noise. See image I just posted to p.b.i.

If I reduce the thread count to 1 then it's only the first top-left 
block that is identical - the rest of the image is random.

Post a reply to this message

Am 28.08.2014 11:12, schrieb scott:

>> That aside, you can always specify a custom seed (or, more precisely,
>> seed base, as UberPOV will still add the thread number to it) using the
>> "+SSn" or "Stochastic_Seed=n" option.
>
> It doesn't recognise the +SS option here (beta 7 release).

Dang. Forget everything I've said; I just notice that I haven't built 
any Windows beta yet that includes those changes; in beta 7, the RNG is 
indeed seeded with a default value. Help is on the way, expect a new 
beta release later tonight.

Post a reply to this message

Am 28.08.2014 18:34, schrieb clipka:
> Am 28.08.2014 11:12, schrieb scott:
>
>>> That aside, you can always specify a custom seed (or, more precisely,
>>> seed base, as UberPOV will still add the thread number to it) using the
>>> "+SSn" or "Stochastic_Seed=n" option.
>>
>> It doesn't recognise the +SS option here (beta 7 release).
>
> Dang. Forget everything I've said; I just notice that I haven't built
> any Windows beta yet that includes those changes; in beta 7, the RNG is
> indeed seeded with a default value. Help is on the way, expect a new
> beta release later tonight.

Done.

Post a reply to this message

>>>> That aside, you can always specify a custom seed (or, more precisely,
>>>> seed base, as UberPOV will still add the thread number to it) using the
>>>> "+SSn" or "Stochastic_Seed=n" option.
>>>
>>> It doesn't recognise the +SS option here (beta 7 release).
>>
>> Dang. Forget everything I've said; I just notice that I haven't built
>> any Windows beta yet that includes those changes; in beta 7, the RNG is
>> indeed seeded with a default value. Help is on the way, expect a new
>> beta release later tonight.
>
> Done.

Excellent thanks :-)

Post a reply to this message