[FFmpeg-devel] [PATCH] Change default behaviour of scale filter from 'progressive' to 'auto'
mark at mdsh.com
Mon Apr 2 23:11:01 CEST 2012
On 02/04/2012 08:43, Tim Nicholson wrote:
> On 31/03/12 01:40, Carl Eugen Hoyos wrote:> Reimar Döffinger<Reimar.Doeffinger<at> gmx.de> writes:> >> Which of course is generally related to any kind of>> "interlaced content" flag being really, really unreliable.>
> I thought it was now much better.....
>> Which is why I am slightly against this patch:> Afaict, on DVB-T (SD), all material is signalled as being> interlaced, although all movies and series are progressive> (because 24Hz stretched and "interlaced" to 50Hz does not> produce visible fields).
> Ahh, the so called PSF mode, (Progressive Sequential Field). The troubleis there is may also be true interlaced material within these streams aswell. In fact within a single program you may find a mix of both typesif the content switches between studio and film based material. So Iwould challenge the "all material" in your statement.
>> I suspect who works with interlaced material knows it> and can set the scaling correctly.>
> Whilst this is true there is one major snafu in the current arrangement,and that is that quite often in a filterchain a scale filter gets autoinserted between other filters, and by default handles materialprogressively. Therefore when building filtergraphs one has to be verycareful to forced add scale filters with interlace enabled at everyplace where a scale filter would otherwise be auto inserted.
> It is much neater, and leads to clearer filtergraphs, if one could forcethe interlace type at the start of a filterchain using 'setfield' andleave the rest of the chain containing just the required filters for theactual required function.
> Handling PSF material as interlaced can mean that non optimalinterpolation is used, but the material is not otherwise damaged.However the reverse of handling interlaced as progressive can rendermaterial unusable..
I've been working with BBC R&B a lot over the last six months trying to
get the best quality we can for HD to SD conversion (with the reason
being that all the BBC Journalism cameras have gone to HD, but most
broadcasts from BBC Journalism are still SD).
The "problem" with lots of interlaced material is that large parts of
the picture can be stationary - effectively progressive.
I'll demonstrate in another email (when I get my act together) the
issues you can get if you scale progressive material with an interlaced
scaler (it's not criticism of the scaler!).
The outcome of our tests showed that you get a better quality scale of
interlaced material if you deinterlace it first, scale it, and
re-interlace it again. But than then gives another problem - the
vertical resolution is too high. Interlaced material should have a
vertical resolution of about 0.68 times that of a progressive image,
otherwise you get horrible vertical twittering.
More information about the ffmpeg-devel