[Bf-cycles] Sample filtering options

Brecht Van Lommel brechtvanlommel at pandora.be
Thu Aug 6 18:33:43 CEST 2015


I strongly suggest to not add any pixel filters with negative weights
to Cycles, like Catmull-rom, Sincor Lanczos. In my opinion they only
add problems and I would not suggest anyone to use them in practice.
Blackman–Harris could be good to add, since that's a filter with
positive weights that supposedly has less ringing than Gaussian. But
it's the only one I can think of.

With a path tracer like Cycles there's enough AA samples that you
don't need to artificially sharpen the image, and deal with problems
like ringing or clamping negative values. I think it's a legacy
feature that is no longer needed. The large filter collections offered
by renderers was never useful in my opinion, it's just needless UI
complexity, most of the filters are just not good.

Also note that denoising algorithms work best with filter importance
sampling since all the pixels are then decorrelated. So if a denoiser
is implemented at some point it would be simplest if it just worked
with all filters.


On Thu, Aug 6, 2015 at 5:12 PM, Nathan Vegdahl <cessen at cessen.com> wrote:
> Yes, I am talking about pixel/sample/AA filters.  Look up the paper "Filter
> Importance Sampling" for more information.
>
> As for texture filtering, I'm sure there's more flexibility there, as I'm
> pretty sure it uses more standard techniques.
>
> --Nathan
>
> On Aug 6, 2015 8:08 AM, "Yury Baranov" <cucumberer at gmail.com> wrote:
>>
>> Hi,
>> I was talking about other filters (maybe they called AA filters). There
>> are two of them ATM: Box and Gaussian. And Bitmap filters are Smart, Cubic,
>> Closest and Linear.
>>
>> 2015-08-06 18:01 GMT+03:00 Nathan Vegdahl <cessen at cessen.com>:
>>>
>>> IIRC Cycles uses "filter importance sampling" to do its sample filtering.
>>> This has some advantages, but one of its disadvantages is that it doesn't
>>> work well with filters that have negative lobes (the resulting renders are
>>> very noisy).
>>>
>>> All the filters you mention have negative lobes, so although they
>>> technically could be implemented, my guess is that they won't be.  It would
>>> be very unexpected behavior to a user if they switched filters and suddenly
>>> their image was twice as noisey.
>>>
>>> --Nathan
>>>
>>> On Aug 6, 2015 7:38 AM, "Yury Baranov" <cucumberer at gmail.com> wrote:
>>>>
>>>> Hi. Is it planned to make more sample filtering options like
>>>> catmull-rom, sinc, lanczos etc.? Maybe more filtering options for bitmaps
>>>> will be good, too.
>>>>
>>>> _______________________________________________
>>>> Bf-cycles mailing list
>>>> Bf-cycles at blender.org
>>>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>>>
>>>
>>> _______________________________________________
>>> Bf-cycles mailing list
>>> Bf-cycles at blender.org
>>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>>
>>
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>
>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
>


More information about the Bf-cycles mailing list