diff mbox series

[FFmpeg-devel,MXF] - Set aspect ratio for jpeg2000 images

Message ID 5bdcb479-65d6-4845-a443-eb66f517427b@ektacom.com
State New
Headers show
Series [FFmpeg-devel,MXF] - Set aspect ratio for jpeg2000 images | expand

Checks

Context Check Description
andriy/configure_x86 warning Failed to apply patch
yinshiyou/configure_loongarch64 warning Failed to apply patch

Commit Message

Cédric Le Barz Jan. 26, 2024, 6:02 p.m. UTC
Set aspect ratio for jpeg2000 images.
Signed-off-by: Cedric Le Barz <clebarz@ektacom.com>
---
 ffmpeg/libavformat/mxfenc.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

Comments

Pierre-Anthony Lemieux Jan. 26, 2024, 6:12 p.m. UTC | #1
On Fri, Jan 26, 2024 at 10:02 AM Cédric Le Barz <clebarz@ektacom.com> wrote:
>
> Set aspect ratio for jpeg2000 images.

"The image area on the reference grid is defined by its upper left
hand reference grid point at location (XOsiz, YOsiz), and its lower
right hand reference grid point at location (Xsiz – 1, Ysiz – 1)"
(15444-1, 8.2)

AFAIK setting the MXF aspect ratio item to the ratio of the image
pixel dimensions only works if (a) the pixel aspect ratio is 1 and (b)
the video is not interlaced -- see 6.2.1.2 and Annex H at
https://pub.smpte.org/doc/st2067-21/20221124-pub/.


> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
Tomas Härdin Jan. 30, 2024, 9:47 a.m. UTC | #2
fre 2024-01-26 klockan 10:12 -0800 skrev Pierre-Anthony Lemieux:
> On Fri, Jan 26, 2024 at 10:02 AM Cédric Le Barz <clebarz@ektacom.com>
> wrote:
> > 
> > Set aspect ratio for jpeg2000 images.
> 
> "The image area on the reference grid is defined by its upper left
> hand reference grid point at location (XOsiz, YOsiz), and its lower
> right hand reference grid point at location (Xsiz – 1, Ysiz – 1)"
> (15444-1, 8.2)
> 
> AFAIK setting the MXF aspect ratio item to the ratio of the image
> pixel dimensions only works if (a) the pixel aspect ratio is 1 and
> (b)
> the video is not interlaced -- see 6.2.1.2 and Annex H at
> https://pub.smpte.org/doc/st2067-21/20221124-pub/.

You also need to know the display rectangle. If for example VBI lines
are present then these might also need be cropped, in addition to the
cropping done at the J2K level. Sometimes container and essence
disagree on what SAR is, which affects the DAR. Which value is the
correct one typically depends on business logic

A demuxer is the wrong place to put logic like this

/Tomas
Tomas Härdin Jan. 30, 2024, 10:36 a.m. UTC | #3
tis 2024-01-30 klockan 10:47 +0100 skrev Tomas Härdin:
> fre 2024-01-26 klockan 10:12 -0800 skrev Pierre-Anthony Lemieux:
> > On Fri, Jan 26, 2024 at 10:02 AM Cédric Le Barz
> > <clebarz@ektacom.com>
> > wrote:
> > > 
> > > Set aspect ratio for jpeg2000 images.
> > 
> > "The image area on the reference grid is defined by its upper left
> > hand reference grid point at location (XOsiz, YOsiz), and its lower
> > right hand reference grid point at location (Xsiz – 1, Ysiz – 1)"
> > (15444-1, 8.2)
> > 
> > AFAIK setting the MXF aspect ratio item to the ratio of the image
> > pixel dimensions only works if (a) the pixel aspect ratio is 1 and
> > (b)
> > the video is not interlaced -- see 6.2.1.2 and Annex H at
> > https://pub.smpte.org/doc/st2067-21/20221124-pub/.
> 
> You also need to know the display rectangle. If for example VBI lines
> are present then these might also need be cropped, in addition to the
> cropping done at the J2K level. Sometimes container and essence
> disagree on what SAR is, which affects the DAR. Which value is the
> correct one typically depends on business logic
> 
> A demuxer is the wrong place to put logic like this

Wups, missed that this is a muxer patch, not a demuxer one.

Anyway, this patch is very wrong. DAR is a container level thing. We
shouldn't take it from the essence unless there is overwhelming reason
to do so. My point about VBI still stands, in modified form: if VBI
lines are present then this patch would write a bogus DAR

Much of the logic currently in mxfenc.c should probably live somewhere
higher up since MOV also supports some of these things

/Tomas
diff mbox series

Patch

diff --git a/ffmpeg/libavformat/mxfenc.c b/ffmpeg/libavformat/mxfenc.c
index 0f13068..7ea19ca 100644
--- a/ffmpeg/libavformat/mxfenc.c
+++ b/ffmpeg/libavformat/mxfenc.c
@@ -2639,7 +2639,6 @@  static int mxf_parse_jpeg2000_frame(AVFormatContext *s, AVStream *st, AVPacket *
 {
     MXFContext *mxf = s->priv_data;
     MXFStreamContext *sc = st->priv_data;
-    AVIOContext *pb = s->pb;
     int component_count = av_pix_fmt_count_planes(st->codecpar->format);
     GetByteContext g;
     uint32_t j2k_ncomponents;
@@ -2679,6 +2678,10 @@  static int mxf_parse_jpeg2000_frame(AVFormatContext *s, AVStream *st, AVPacket *
     }
     bytestream2_get_bufferu(&g, sc->j2k_info.j2k_comp_desc, 3 * j2k_ncomponents);
 
+    /* Set aspect ratio */
+    av_reduce(&sc->aspect_ratio.num, &sc->aspect_ratio.den,
+              sc->j2k_info.j2k_xsiz, sc->j2k_info.j2k_ysiz, INT32_MAX);
+
     sc->frame_size = pkt->size;
 
     return 1;