[FFmpeg-user] Objectively best deinterlacer?

[Robert Krüger]

On Wed, Jan 22, 2014 at 4:59 PM, Leo Izen <leo.izen at gmail.com> wrote:
> On 01/22/2014 05:33 AM, Paul B Mahol wrote:
>>
>> On 1/22/14, Leo Izen <leo.izen at gmail.com> wrote:
>>>
>>> -filter:v 'w3fdif,select=outputs=2:expr=mod(n\,2)+1,blend'
>>
>>
>> This does nothing as blend with no options returns first input.
>
>
> I tried using blend=all_expr=0.5*(A+B) and the results looked much better,
> but now it encoded at an annoying 7 fps (and this is mostly filter overhead;
> I was encoding to ffv1 which is superfast)
>
>
>
> On 01/22/2014 10:00 AM, Phil Rhodes wrote:
>> I've used Yadif with great success, including critical things like camera
>> originals shot interlaced which were needed as progressive.
>>
>> I'm not sure what techniques it uses, but I suspect it uses some adaptive
>> stuff to avoid blurring stationary objects.
>>
>> I would say it is subjectively as good as some of the expensive boxes that
>> do deinterlacing.
>>
>> P
>
> Compared with w3fdif + blend, I looked at yadif and decided that it was
> sharper (and also filtered 3x faster). I haven't done much blind comparison
> to -filter:v pp=fd though (AFAIK this one was the traditional filter used
> with -deinterlace). I'll have to do more subjective tests.

why do you use blend with w3fdif and not just discard every other
frame if you're deinterlacing 60i->30p by using the -r output option?
I don't see what you gain by blending in terms of quality.
Regarding yadif I have to second what Phil wrote. I have used it a lot
myself and never had a quality problem and I know of at least one
large media organization that has been using yadif massively for years
after thorough analysis and comparison with commercial options. You
can search the archives for postings on w3fdif. there were some
comments from the original author of the filter from the BBC who
described some points where w3fdif performed better (in terms of
quality) than yadif in some of his tests.