[FFmpeg-devel] [PATCH] avcodec/h264_parser: remove key frame tagging heuristics

[James Almer]

On 7/12/2021 11:44 PM, Kieran Kunhya wrote:
> On Tue, 13 Jul 2021, 02:45 James Almer, <jamrial at gmail.com> wrote:
> 
>> On 7/12/2021 10:01 PM, Kieran Kunhya wrote:
>>>>
>>>> Because it isn't something that should be marked as a keyframe as coded
>>>> bitstream in any kind of container, like it's the case of mp4 sync
>> samples.
>>>>
>>>
>>> MPEG-TS Random Access Indicator expects keyframes to be signalled like
>> this.
>>> With intra-refresh and this code removed, there will be no random access
>>> points at all.
>>
>> If MPEG-TS wants to tag packets containing things other than IDR access
>> units as RAPs, then it should analyze the bitstream itself in order to
>> tag them itself as such in the output.
>> This parser as is is generating invalid output for other containers that
>> are strict about key frames, and signal recovery points (like those
>> indicated by the use of this SEI) by other means.
>>
> 
> Why not just detect IDR in containers that only care about that (which is a
> mistake because if things like open gop)? Doing that's is relatively simple
> compared to adding bitstream parsing into MPEGTS.

Both would require bitstream parsing in a muxer to look at what pictures 
are present within the AU in the packet, so I'm inclined to do it for 
the muxer that wants to tag more things than normal as keyframes instead 
of in every other muxer.

I will keep tagging these packets as keyframes in the parser but behind 
a new flag that's disabled by default, then insert an internal instance 
of the parser to mpegtsenc with said flag enabled, and use that to mark 
mpegts RAPs.
This has the added benefit of marking these packets as RAPs in mpegts on 
codec copy scenarios when the source is an mp4 or mkv and the generic 
code never uses an AVParser.

> 
> Kieran
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>