mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-08-04 04:35:06 +00:00
Code Issues 1 Projects Releases Wiki Activity

rtph264depay: only guess AU boundaries when aren't indicated by marker

Browse source 
The marker bit isn't mandatory and we had in place code to guess AU
boundaries by detecting a new picture start. This guessing code
didn't work with interlaced content that has proper marker bits
to indicate the AU boundaries. It was leaking the first field buffer
and producing a corrupted output.

fixes: https://bugzilla.gnome.org/show_bug.cgi?id=728041
This commit is contained in:
Josep Torra 2014-04-11 18:19:49 +02:00 committed by Wim Taymans
parent b32de2794b
commit eaee14aff4
1 changed files with 26 additions and 20 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
gst/rtp/gstrtph264depay.c
View file

@ -712,29 +712,35 @@ gst_rtp_h264_depay_handle_nal (GstRtpH264Depay * rtph264depay, GstBuffer * nal,
if (rtph264depay->merge) {
gboolean start = FALSE, complete = FALSE;
/* consider a coded slices (IDR or not) to start a picture,
* (so ending the previous one) if first_mb_in_slice == 0
* (non-0 is part of previous one) */
/* NOTE this is not entirely according to Access Unit specs in 7.4.1.2.4,
* but in practice it works in sane cases, needs not much parsing,
* and also works with broken frame_num in NAL (where spec-wise would fail) */
if (nal_type == 1 || nal_type == 2 || nal_type == 5) {
/* we have a picture start */
start = TRUE;
if (map.data[5] & 0x80) {
/* first_mb_in_slice == 0 completes a picture */
/* marker bit isn't mandatory so in the following code we try to guess
* an AU boundary by detecting a new picture start */
if (!marker) {
/* consider a coded slices (IDR or not) to start a picture,
* (so ending the previous one) if first_mb_in_slice == 0
* (non-0 is part of previous one) */
/* NOTE this is not entirely according to Access Unit specs in 7.4.1.2.4,
* but in practice it works in sane cases, needs not much parsing,
* and also works with broken frame_num in NAL (where spec-wise would fail) */
/* FIXME: this code isn't correct for interlaced content as AUs should be
* constructed with pairs of fields and the guess here will just push out
* AUs with a single field in it */
if (nal_type == 1 || nal_type == 2 || nal_type == 5) {
/* we have a picture start */
start = TRUE;
if (map.data[5] & 0x80) {
/* first_mb_in_slice == 0 completes a picture */
complete = TRUE;
}
} else if (nal_type >= 6 && nal_type <= 9) {
/* SEI, SPS, PPS, AU terminate picture */
complete = TRUE;
}
} else if (nal_type >= 6 && nal_type <= 9) {
/* SEI, SPS, PPS, AU terminate picture */
complete = TRUE;
GST_DEBUG_OBJECT (depayload, "start %d, complete %d", start, complete);
if (complete && rtph264depay->picture_start)
outbuf = gst_rtp_h264_complete_au (rtph264depay, &out_timestamp,
&out_keyframe);
}
GST_DEBUG_OBJECT (depayload, "start %d, complete %d", start, complete);
if (complete && rtph264depay->picture_start)
outbuf = gst_rtp_h264_complete_au (rtph264depay, &out_timestamp,
&out_keyframe);
/* add to adapter */
gst_buffer_unmap (nal, &map);