| Message ID | 20220410181200.3181-1-cus@passwd.hu |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel,1/2] avformat/mov: fix timecode with high frame rate content | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
Quoting Marton Balint (2022-04-10 20:11:59) > 60 fps content have "Number of Frames" set to 30 in the tmcd atom, but the > frame duration / timescale reflects the original video frame rate. > > Therefore we multiply the frame count with the quotient of the rounded timecode > frame rate and the "Number of Frames" per second to get a frame count in the original > (higher) frame rate. > > Note that the frames part in the timecode will be in high frame rate which will > make the timecode different to e.g. MediaInfo which seems to show the 30 fps > timecode even for 120 fps content. > > Regression since 428b4aacb1a91a267650de644519882a5f700388. > > Fixes ticket #9710. > Fixes ticket #9492. Sounds like there should be a test for this.
On Mon, 11 Apr 2022, Anton Khirnov wrote: > Quoting Marton Balint (2022-04-10 20:11:59) >> 60 fps content have "Number of Frames" set to 30 in the tmcd atom, but the >> frame duration / timescale reflects the original video frame rate. >> >> Therefore we multiply the frame count with the quotient of the rounded timecode >> frame rate and the "Number of Frames" per second to get a frame count in the original >> (higher) frame rate. >> >> Note that the frames part in the timecode will be in high frame rate which will >> make the timecode different to e.g. MediaInfo which seems to show the 30 fps >> timecode even for 120 fps content. >> >> Regression since 428b4aacb1a91a267650de644519882a5f700388. >> >> Fixes ticket #9710. >> Fixes ticket #9492. > > Sounds like there should be a test for this. The smallest file I managed to find which is affected by this is mov/canon_6d/mvi_9114.mov, but that is still 12 MB, therefore probably not fit for addition to fate-samples. With our muxer, the issue is not reproducible, so remuxing is not an option. Regards, Marton
On Mon, 11 Apr 2022, Marton Balint wrote: > > > On Mon, 11 Apr 2022, Anton Khirnov wrote: > >> Quoting Marton Balint (2022-04-10 20:11:59) >>> 60 fps content have "Number of Frames" set to 30 in the tmcd atom, but >>> the >>> frame duration / timescale reflects the original video frame rate. >>> >>> Therefore we multiply the frame count with the quotient of the rounded >>> timecode >>> frame rate and the "Number of Frames" per second to get a frame count in >>> the original >>> (higher) frame rate. >>> >>> Note that the frames part in the timecode will be in high frame rate >>> which will >>> make the timecode different to e.g. MediaInfo which seems to show the 30 >>> fps >>> timecode even for 120 fps content. >>> >>> Regression since 428b4aacb1a91a267650de644519882a5f700388. >>> >>> Fixes ticket #9710. >>> Fixes ticket #9492. >> >> Sounds like there should be a test for this. > > The smallest file I managed to find which is affected by this is > mov/canon_6d/mvi_9114.mov, but that is still 12 MB, therefore probably not > fit for addition to fate-samples. > > With our muxer, the issue is not reproducible, so remuxing is not an > option. Will apply soon. Regards, Marton
diff --git a/libavformat/isom.h b/libavformat/isom.h index 5caf42b15d..99408a42d1 100644 --- a/libavformat/isom.h +++ b/libavformat/isom.h @@ -214,6 +214,7 @@ typedef struct MOVStreamContext { int has_palette; int64_t data_size; uint32_t tmcd_flags; ///< tmcd track flags + uint8_t tmcd_nb_frames; ///< tmcd number of frames per tick / second int64_t track_end; ///< used for dts generation in fragmented movie files int start_pad; ///< amount of samples to skip due to enc-dec delay unsigned int rap_group_count; diff --git a/libavformat/mov.c b/libavformat/mov.c index 6c847de164..8527591ff3 100644 --- a/libavformat/mov.c +++ b/libavformat/mov.c @@ -2364,6 +2364,7 @@ static int mov_parse_stsd_data(MOVContext *c, AVIOContext *pb, tmcd_ctx->tmcd_flags = val; st->avg_frame_rate.num = AV_RB32(st->codecpar->extradata + 8); /* timescale */ st->avg_frame_rate.den = AV_RB32(st->codecpar->extradata + 12); /* frameDuration */ + tmcd_ctx->tmcd_nb_frames = st->codecpar->extradata[16]; /* number of frames */ if (size > 30) { uint32_t len = AV_RB32(st->codecpar->extradata + 18); /* name atom length */ uint32_t format = AV_RB32(st->codecpar->extradata + 22); @@ -7893,11 +7894,16 @@ static int mov_read_timecode_track(AVFormatContext *s, AVStream *st) FFStream *const sti = ffstream(st); int flags = 0; int64_t cur_pos = avio_tell(sc->pb); - uint32_t value; + int64_t value; + AVRational tc_rate = st->avg_frame_rate; + int rounded_tc_rate; if (!sti->nb_index_entries) return -1; + if (!tc_rate.num || !tc_rate.den || !sc->tmcd_nb_frames) + return -1; + avio_seek(sc->pb, sti->index_entries->pos, SEEK_SET); value = avio_rb32(s->pb); @@ -7910,6 +7916,12 @@ static int mov_read_timecode_track(AVFormatContext *s, AVStream *st) * No sample with tmcd track can be found with a QT timecode at the moment, * despite what the tmcd track "suggests" (Counter flag set to 0 means QT * format). */ + + /* 60 fps content have tmcd_nb_frames set to 30 but tc_rate set to 60, so + * we multiply the frame number with the quotient. */ + rounded_tc_rate = (tc_rate.num + tc_rate.den / 2) / tc_rate.den; + value = av_rescale(value, rounded_tc_rate, sc->tmcd_nb_frames); + parse_timecode_in_framenum_format(s, st, value, flags); avio_seek(sc->pb, cur_pos, SEEK_SET);
60 fps content have "Number of Frames" set to 30 in the tmcd atom, but the frame duration / timescale reflects the original video frame rate. Therefore we multiply the frame count with the quotient of the rounded timecode frame rate and the "Number of Frames" per second to get a frame count in the original (higher) frame rate. Note that the frames part in the timecode will be in high frame rate which will make the timecode different to e.g. MediaInfo which seems to show the 30 fps timecode even for 120 fps content. Regression since 428b4aacb1a91a267650de644519882a5f700388. Fixes ticket #9710. Fixes ticket #9492. Signed-off-by: Marton Balint <cus@passwd.hu> --- libavformat/isom.h | 1 + libavformat/mov.c | 14 +++++++++++++- 2 files changed, 14 insertions(+), 1 deletion(-)