Le 16-08-07 à 15:31, Stephen a écrit :
> On 8/7/2016 8:12 PM, Alain wrote:
>>>
>>
>> In method 1 and 2, confidence, variance and ratio are used to control
>> how many samples are to be used in the different intervals depending on
>> the density and illumination in the various intervals. With method 3,
>> using more that a single interval is highly counterproductive.
>>
>> I've done my tests with intervals and the negative impact they have on
>> the performance is astounding.
>> I demonstrated that samples 100 is way faster than intervals 10 samples
>> 1 for greatly better results.
>
>
> Thanks Alain, I knew I would get it in the neck for posting that code. :)
> Except for the intervals/samples thing. The other things you mention are
> Bishop3D defaults and there is not a lot that I can do about what is
> exported except go through the code afterwards. But I have a relatively
> fast machine, so I don't.
> I seem to have a blind spot for samples vs intervals. Thomas, amongst
> others, is forever mentioning it.
> I use media so seldom that I have to relearn it whenever I want to use
> it. :(
> So it is not ignorance but stupidity on my part. :(
>

Here are the default vbalues :
method 3
samples 10
intervals 1 // Don't increase !

The default code from Bishop mostly repeat the actual defaults values. 
That effectively make then into "no operation" code.

Even on the fastest machine, using intervals >1 is a serious performance 
killer.


Alain

Post a reply to this message