[FFmpeg-devel] [PATCH] avfilter: add framerate video filter
cus at passwd.hu
Sat Aug 29 17:39:28 CEST 2015
On Sat, 29 Aug 2015, Robert Krüger wrote:
> On Sat, Aug 29, 2015 at 4:45 PM, Robert Krüger <krueger at lesspain.de> wrote:
>> On Sat, Aug 29, 2015 at 3:53 PM, Carl Eugen Hoyos <cehoyos at ag.or.at>
>>> Robert Krüger <krueger <at> lesspain.de> writes:
>>> > Do you mean processing the top and bottom fields
>>> > as separate progressive
>>> (Sorry, I thought this is what "deinterleaving and
>>> interleaving" means.)
>>> > streams?
>>> Not necessarily "streams", just not processing them
>>> interlaced but deinterleaved.
>>> If this leads to shadows, then I would suggest to
>>> use two framerate instances.
>> I understand the variant with processing the fields as separate streams
>> with two framerate instances, but not the other one. One would have to test
>> but the more I think about it, I think it could very well work and result
>> in better quality as the deinterlacing/reinterlacing obviously destroys
>> motion information.
>> One would probably have to make tests with interlaced footage that
>> contains a lot of motion.
>> It would certainly be interesting to see that filter chain, though. I
>> haven't worked with the il filter at all, so I would have to read up on
>> that first.
> Which I should have done first. I read the il filter docs and now I
> understand what you mean. If I am correct in assuming, what the framerate
> filter does, this might work just with one stream and would do what I
> described as using two streams.
I am afraid I still don't understand how a non-deinterlacing approach
can give better quality. Without deinterlacing you will blend frames which
are 2 units distant in time, with deinterlacing you blend frames which are
1 unit distant.
Consider you have a 25i source, and you want 30i (60p for display)
Are you referring to this filter chain for deinterleaving?
And this for deinterlacing?
The second is much, much better.
More information about the ffmpeg-devel