[FFmpeg-devel] [PATCH v1] avcodec/h264_metadata_bsf: Fix user data failed to insert in case no SPSs NAL for global headers

[Mark Thompson]

On 26/12/2019 00:55, lance.lmwang at gmail.com wrote:
> From: Limin Wang <lance.lmwang at gmail.com>
> 
> FLV, MP4... will enable global_header default and place SPSs headers in extradata
> instead of every keyframe. So it'll failed to insert user data unregisted
> for no SPSs NAL after first AU without the patch.
> 
> Please test it with below command:
> ./ffmpeg -f lavfi -i testsrc -c:v libx264 -g 25 \
>     -bsf:v h264_metadata=sei_user_data=086f3693-b7b3-4f2c-9653-21492feee5b8+hello \
>     -frames:v 150 test.mp4
> 
> After applied the patch, you'll get the user data for every keyframe with below command:
> ./ffmpeg -i test.mp4 -vf showinfo -frames:v 150 -f null -
> 
> Signed-off-by: Limin Wang <lance.lmwang at gmail.com>
> ---
>  libavcodec/h264_metadata_bsf.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/libavcodec/h264_metadata_bsf.c b/libavcodec/h264_metadata_bsf.c
> index 5de74be9d6..9690ca433b 100644
> --- a/libavcodec/h264_metadata_bsf.c
> +++ b/libavcodec/h264_metadata_bsf.c
> @@ -279,7 +279,7 @@ static int h264_metadata_filter(AVBSFContext *bsf, AVPacket *pkt)
>  {
>      H264MetadataContext *ctx = bsf->priv_data;
>      CodedBitstreamFragment *au = &ctx->access_unit;
> -    int err, i, j, has_sps;
> +    int err, i, j, has_sps, is_keyframe = 0;
>      H264RawAUD aud;
>  
>      err = ff_bsf_get_packet_ref(bsf, pkt);
> @@ -359,11 +359,13 @@ static int h264_metadata_filter(AVBSFContext *bsf, AVPacket *pkt)
>                  goto fail;
>              has_sps = 1;
>          }
> +        if (au->units[i].type == H264_NAL_IDR_SLICE)
> +            is_keyframe = 1;
>      }
>  
>      // Only insert the SEI in access units containing SPSs, and also
>      // unconditionally in the first access unit we ever see.
> -    if (ctx->sei_user_data && (has_sps || !ctx->done_first_au)) {
> +    if (ctx->sei_user_data && (has_sps || !ctx->done_first_au || is_keyframe)) {
>          H264RawSEIPayload payload = {
>              .payload_type = H264_SEI_TYPE_USER_DATA_UNREGISTERED,
>          };
> 

This was avoided originally because identifying key frames (or really, seek points) in an H.264 stream is really hard.  I don't particularly like the idea of having it work in some streams but not others, depending on exactly how the stream was encoded - hence the original decision to only include it in access units which were already distinguished as flat-file seek-points by having the parameter sets in them.

I guess I'm not against this in itself, but please add more of a comment in the code explaining the rationale and noting the cases which are not included.  (And if you can cover more cases, such as recovery points, then that would be even better.)

Thanks,

- Mark