[FFmpeg-devel] [PATCH] Handle AVID MJPEG streams directly in the MJPEG decoder.

[James Almer]

On 2/27/2021 1:37 PM, James Almer wrote:
> On 2/27/2021 1:33 PM, Michael Niedermayer wrote:
>> On Wed, Feb 24, 2021 at 11:25:42AM +0100, Anton Khirnov wrote:
>>> AVID streams - currently handled by the AVRN decoder - can be (depending
>>> on extradata contents) either MJPEG or raw video. To decode the MJPEG
>>> variant, the AVRN decoder currently instantiates a MJPEG decoder
>>> internally and forwards decoded frames to the caller (possibly after
>>> cropping them).
>>>
>>> This is suboptimal, because the AVRN decoder does not forward all the
>>> features of the internal MJPEG decoder, such as direct rendering.
>>> Handling such forwarding in a full and generic manner would be quite
>>> hard, so it is simpler to just handle those streams in the MJPEG decoder
>>> directly.
>>>
>>> The AVRN decoder, which now handles only the raw streams, can now be
>>> marked as supporting direct rendering.
>>>
>>> This also removes the last remaining internal use of the obsolete
>>> decoding API.
>>> ---
>>> And now without the missing isom tag change. Thanks to Andreas for 
>>> noticing.
>>> ---
>>>   configure               |  1 -
>>>   libavcodec/avrndec.c    | 71 +----------------------------------------
>>>   libavcodec/mjpegdec.c   | 11 +++++++
>>>   libavcodec/version.h    |  2 +-
>>>   libavformat/avidec.c    |  6 ++++
>>>   libavformat/isom_tags.c |  4 +--
>>>   tests/fate/video.mak    |  2 +-
>>>   7 files changed, 22 insertions(+), 75 deletions(-)
>>
>> breaks:
>> ./ffmpeg -vlowres 2 -i ~/tickets/162/avid.avi  -vframes 3 avrn.avi
>>
>> The outputed image is several times bigger and looks as if it encodes 
>> random bits od memory
> 
> Can you try again with git head? I removed the max_lowres value since 
> the avrn raw decoder doesn't support it, since it was pulled from the 
> mjpeg version.

Ok, the above was unrelated since this sample is mjpeg, now raw.
The following however fixes it for me:

> diff --git a/libavcodec/mjpegdec.c b/libavcodec/mjpegdec.c
> index 79c7d6e6cf..8eed8eefe6 100644
> --- a/libavcodec/mjpegdec.c
> +++ b/libavcodec/mjpegdec.c
> @@ -2863,8 +2863,8 @@ the_end:
>      if ((avctx->codec_tag == MKTAG('A', 'V', 'R', 'n') ||
>           avctx->codec_tag == MKTAG('A', 'V', 'D', 'J')) &&
>          avctx->coded_height > s->orig_height) {
> -        frame->height   = avctx->coded_height;
> -        frame->crop_top = frame->height - s->orig_height;
> +        frame->height   = AV_CEIL_RSHIFT(avctx->coded_height, avctx->lowres);
> +        frame->crop_top = frame->height - AV_CEIL_RSHIFT(s->orig_height, avctx->lowres);

But it may be an incomplete solution, since there's another similar 
check/assignment in ff_mjpeg_decode_sof() also added by 9e4225cf7f.