[FFmpeg-devel] [RCF] lavfi aspect ratio setting path

[Baptiste Coudurier]

Hi Stefano,

On 11/20/10 3:42 AM, Stefano Sabatini wrote:
> On date Friday 2010-11-19 19:14:28 -0800, Baptiste Coudurier encoded:
>> On 11/19/2010 06:55 PM, Michael Niedermayer wrote:
>>> On Thu, Nov 18, 2010 at 04:45:48PM -0800, Baptiste Coudurier wrote:
>>>> On 11/18/2010 04:33 PM, Michael Niedermayer wrote:
>>>>> On Thu, Nov 18, 2010 at 04:12:29PM -0800, Baptiste Coudurier wrote:
>>>>>> Hi Stefano,
>>>>>>
>>>>>> On 11/18/2010 01:27 PM, Stefano Sabatini wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> -aspect currently works by changing the PAR in the video encoder
>>>>>>> context and in the output stream, thus readjusting the output DAR.
>>>>>>>
>>>>>>> If the libavfilter is enabled, in output_packet() we have:
>>>>>>>       if (ist->picref->video)
>>>>>>>           ost->st->codec->sample_aspect_ratio = ist->picref->video->pi
>>>>>>>
>>>>>>> so sample_aspect_ratio is set to the value in the filtered frame,
>>>>>>> thus ignoring the value set with -aspect.
>>>>>>>
>>>>>>> Workaround for this issue is either configure with --disable-avfilter
>>>>>>> and use -aspect, or use the aspect filter.
>>>>>>>
>>>>>>> I suggest to disable the -aspect option when using libavfilter, thus
>>>>>>> forcing the users to employ the aspect filter. Alternatively we could
>>>>>>> just drop support for -aspect in favor of libavfilter, but when that
>>>>>>> approach made sense with the pad and crop options which were
>>>>>>> "polluting" the ffmpeg code with much hard-to-maintain complex code,
>>>>>>> this doesn't seem the case for the aspect option and I still want to
>>>>>>> keep that option for users which for some reason (e.g. because of the
>>>>>>> lavfi regressions) are using an ffmpeg compiled with
>>>>>>> --disable-avfilter.
>>>>>>
>>>>>> First, thanks for looking into this.
>>>>>>
>>>>>> I, too, would like to keep the -aspect cli option.
>>>>>>
>>>>>> I have a patch in my tree that actually set the aspect at the end of
>>>>>> configure_filters and disable the code you quoted above.
>>>>>
>>>>> can you post that to the list so i can take a look?
>>>> >from your description it sounds like a good solution
>>>>
>>>> Here it is.
>>>>
>>>> --
>>>> Baptiste COUDURIER
>>>> Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
>>>> FFmpeg maintainer                                  http://www.ffmpeg.org
>>>
>>>>  ffmpeg.c |    7 +++----
>>>>  1 file changed, 3 insertions(+), 4 deletions(-)
>>>> 52f66ace0baf991b397cbe489f8df5b7623e0f33  0001-Fix-output-aspect-ratio-with-libavfilter.patch
>>>> From 246c5cdba7defcd697597549fd110fb8b86cae62 Mon Sep 17 00:00:00 2001
>>>> From: bcoudurier<baptiste.coudurier at gmail.com>
>>>> Date: Sun, 24 Oct 2010 22:33:42 -0700
>>>> Subject: [PATCH 1/5] Fix output aspect ratio with libavfilter
>>>>
>>>> ---
>>>>  ffmpeg.c |    7 +++----
>>>>  1 files changed, 3 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/ffmpeg.c b/ffmpeg.c
>>>> index 1681664..f06cc3e 100644
>>>> --- a/ffmpeg.c
>>>> +++ b/ffmpeg.c
>>>> @@ -450,6 +450,9 @@ static int configure_filters(AVInputStream *ist, AVOutputStream *ost)
>>>>
>>>>      codec->width  = ist->output_video_filter->inputs[0]->w;
>>>>      codec->height = ist->output_video_filter->inputs[0]->h;
>>>> +    codec->sample_aspect_ratio =
>>>> +        av_d2q(frame_aspect_ratio*codec->height/codec->width, 255);
>>>> +    ost->st->sample_aspect_ratio = codec->sample_aspect_ratio;
>>>
>>> this looks like it would break the aspect filter, in the sense that its
>>> value would be ignored. But maybe i miss something?
>>
>> Yes, it will be ignored.
>> The aspect filter handling is weird anyway, dump_format and thus
>> ffmpeg won't display the correct aspect ratio, I consider this
>> broken so I disable his effects.
> 
> The aspect filters have another problem, they may fail with
> some encoder which needs to set the DAR/SAR when openining the
> context, because they set DAR/SAR frame by frame (while this *may* be
> a global property).

I think you are right.

> If this is the case (I need confirmation from someone with more
> expertise in encoders), then the only way to fix the problem is to
> make the aspect a property of the link, and make it configurable
> during the configuration stage. In case the aspect changes during
> filtering, we may need to re-configure the filtergraph (as it should
> happen when w/h change).

That may be needed.

In the mean time, I suggest we at least restore the -aspect cli option
functionality, so users can have something working, so I believe my
patch could be acceptable.

-- 
Baptiste COUDURIER
Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer                                  http://www.ffmpeg.org