On 09/05/2024 06.07, Yann Cainjo wrote:
Hi Mark
Hey Yann
DTSes & PTSes only exist in MPEG PESes for I-frames
Where did you get this information ?
By parsing actual presentation streams. Start reading 24 lines down in this message. Note that email
will wrap long sentences but its easy to fix them if you look.
I was reading this article but it seems to me that P and B frames have PTS
and DTS...
Understanding Timelines within MPEG Standards
https://ir.cwi.nl/pub/23650/23650B.pdf
Thank you for that, but it does not resolve the issues.
Best regards
Yann
@@@
Yann CAINJO
Photographie, photojournalisme, vidéo, prise de son, podcast & vidéo 360 VR
Formateur en photographie et audiovisuel
06 73 23 73 31
y...@yanncainjo.com
http://www.yanncainjo.com
Do any of these 'sound' familiar?
A CFR source makes a target that's slightly VFR.
Non-monotonous DTSes.
Concatenations produce a playback glitch or fail altogether.
Use cases that used to work don't work anymore because someone patched the code to cover a different
use case.
I believe they all have the cause: Open GOPs in which open B-frames are mishandled. Here are two
pertinent facts:
1, The exact same video can be in any of four physical packet orders:
..BPBPBBPIBPBP.. //closed GOP, physical stream in PTS order
..PBPBPBBIPBPB.. //closed GOP, physical stream in DTS order
..BPBPBBIBPBP.. //open GOP, physical stream in PTS order
..PBPBIBBPBPB.. //open GOP, physical stream in DTS order
I have seen _all_four_ in _professional_ videos (presumably because MPEG does not specify what an
MPEG video's physical packet order should be).
Note that I've only seen physical packet order that is either PTS order or DTS order but I imagine
anything is possible.
DTS order makes the most sense for minimizing buffers, but strangely, most
videos are in PTS order.
2, MPEG tags (from H.262 §6.2.2.6 & §6.3.8) in the GOP header to handle open
GOPs:
'closed_gop' = '1/0'
'broken_link' = '1/0' //'1': GOP has been edited and B-frames have lost their
reference frames.
These flags are in an extention created to force players to ignore the open
B-frames.
FFmpeg does _nothing_ with these tags when you cut an open GOP.
It's helpful to show DTS order and PTS order as though they are two separate
(imaginary) streams.
They're not two separate streams of course. There's only one stream and DTS &
PTS are just numbers.
A closed GOP with physical order: ..BPBPBBPIBPBP..
time ———>
PTSes .. B P B P B B P I B P B P .. <== physical order
_¦_/ _¦_/ _¦__¦_/ / _¦_/ _¦_/
/ ¦ / ¦ / ¦ ¦ / / ¦ / ¦
DTSes P B P B P B B I P B P B
/
vdcut
A closed GOP with physical order: ..PBPBPBBIPBPB..
time ———>
PTSes B P B P B B P I B P B P
_¦_/ _¦_/ _¦__¦_/ / _¦_/ _¦_/
/ ¦ / ¦ / ¦ ¦ / / ¦ / ¦
DTSes .. P B P B P B B I P B P B .. <== physical order
/
vdcut
An open GOP with physical order: ..BPBPBBIBPBP..
vpcut
time ———> /
PTSes .. B P B P B B I B P B P .. <== physical order
_¦_/ _¦_/ _¦__¦_/ _¦_/ _¦_/
/ ¦ / ¦ / ¦ ¦ / ¦ / ¦
DTSes P B P B I B B P B P B
/ \______/
vdcut drop open Bs found in vdcut..vpcut
An open GOP with physical order: ..PBPBIBBPBPB..
vpcut
time ———> /
PTSes B P B P B B I B P B P
_¦_/ _¦_/ _¦__¦_/ _¦_/ _¦_/
/ ¦ / ¦ / ¦ ¦ / ¦ / ¦
DTSes .. P B P B I B B P B P B .. <== physical order
/ \______/
vdcut drop open Bs found in vdcut..vpcut
The following are parses of actual, professional videos with references to H.222 & H.262 (where the
format documentation can be found). There's an I-frame followed by a P-frame followed by a B-frame
followed by a GOP header. Note 'PTS_DTS_flags' 'picture_coding_type' 'closed_gop' & 'broken_link'.
Here is an I-frame
------------------------------|
Re: H.222 §2.4.3.6 | PES_HEADER________________
PES video stream 0 | 00 00 01 E0 == == == == ==
PES_packet_length | == == == == 07 EC == == == //= 2028 bytes
PES Header data content flags | == == == == == == 81 C0 ==
| .——————' '——————.
marker bits | 10-- ---- ---- ----
PES_scrambling_control | --00 ---- ---- ----
PES_priority | ---- 0--- ---- ----
data_alignment_indicator | ---- -0-- ---- ----
copyright | ---- --0- ---- ---- //1 =
copyrighted
original_or_copy | ---- ---1 ---- ---- //1 =
original; 0 = copy
PTS_DTS_flags | ---- ---- 11-- ---- //00 = PTS DTS don't exist, 01 =
illegal, 10 = PTS only, 11 = PTS & DTS
ESCR_flag | ---- ---- --0- ---- //elementary
stream clock reference
ES_rate_flag | ---- ---- ---0 ----
DSM_trick_mode_flag | ---- ---- ---- 0--- //Digital
Storage Media trick mode
additional_copy_info_flag | ---- ---- ---- -0--
PES_CRC_flag | ---- ---- ---- --0-
PES_extension_flag | ---- ---- ---- ---0 //<-- error;
'1': pes extension present
PES_header_data_length | == == == == == == == == 0A //= 10 bytes
------------------------------|
Re: H.262 §6.2.3 | PICTURE_HEADER_________
picture_start_code | 00 00 01 00 == == == ==
| == == == == 00 8F FF F8
| .—————————' '—————————————————.
temporal_reference | 0000 0000 10-- ---- ---- ---- ---- ----
picture_coding_type | ---- ---- --00 1--- ---- ---- ---- ---- //'001': I-'picture';
'010': P-'picture'; '011': B-'picture'
vbv_delay | ---- ---- ---- -111 1111 1111 1111 1--- //the number of 90 KHz
clocks from picture_start_code to decoder start (or '111 1111 1111 1111 1' for no delay)
extra_bit_picture | ---- ---- ---- ---- ---- ---- ---- -0-- //'1': followed by 8
bits of content_description_data followed by another extra_bit_picture
padding_bits | ---- ---- ---- ---- ---- ---- ---- --00
------------------------------|
Here is a P-frame
------------------------------|
stnsoft.com/DVD/pes-hdr.html | PES_HEADER________________
video_stream_0 | 00 00 01 E0 == == == == ==
PES_packet_length | == == == == 07 EC == == == //= 2028 bytes
(i.e. [0014..0800])
PES Header data content flags | == == == == == == 81 00 ==
| .——————' '——————.
marker bits | 10-- ---- ---- ----
PES_scrambling_control | --00 ---- ---- ----
PES_priority | ---- 0--- ---- ----
data_alignment_indicator | ---- -0-- ---- ----
copyright | ---- --0- ---- ----
original_or_copy | ---- ---1 ---- ----
PTS_DTS_flags | ---- ---- 00-- ----
ESCR_flag | ---- ---- --0- ---- //elementary
stream clock reference
ES_rate_flag | ---- ---- ---0 ----
DSM_trick_mode_flag | ---- ---- ---- 0---
additional_copy_info_flag | ---- ---- ---- -0--
PES_CRC_flag | ---- ---- ---- --0-
PES_extension_flag | ---- ---- ---- ---0 //1 = pes
extension present
PES_header_data_length | == == == == == == == == 00 //= zero bytes
------------------------------|
Re: H.262 §6.2.3 | PICTURE_HEADER_____________________
picture_start_code | 00 00 01 00 == == == == == == == ==
| == == == == 00 57 FF FB 80 == == ==
| .——————————'
'———————————————————————.
temporal_reference | 0000 0000 01-- ---- ---- ---- ---- ---- ----
----
picture_coding_type | ---- ---- --01 0--- ---- ---- ---- ---- ---- ---- //'001':
I-picture; '010': P-picture; '011': B-picture
vbv_delay | ---- ---- ---- -111 1111 1111 1111 1--- ---- ---- //= no delay;
else, the number of 90 KHz clocks from picture_start_code to decoder start
full_pel_forward_vector | ---- ---- ---- ---- ---- ---- ---- -0-- ----
---- //ignored
forward_f_code | ---- ---- ---- ---- ---- ---- ---- --11 1---
---- //ignored
full_pel_backward_vector | ---- ---- ---- ---- ---- ---- ---- ---- -0--
---- //ignored
backward_f_code | ---- ---- ---- ---- ---- ---- ---- ---- --00
0--- //ignored
extra_bit_picture | ---- ---- ---- ---- ---- ---- ---- ---- ---- -0-- //= no
content_description_data; '1' = followed by 8 bits of content_description_data followed by another
extra_bit_picture
padding_bits | ---- ---- ---- ---- ---- ---- ---- ---- ----
--00
padding bytes | == == == == == == == == == 00 00 00
------------------------------|
Here is a B-frame
------------------------------|
stnsoft.com/DVD/pes-hdr.html | PES_HEADER________________
video_stream_0 | 00 00 01 E0 == == == == ==
PES_packet_length | == == == == 07 EC == == == //= 2028 bytes
(i.e. [0014..0800])
PES Header data content flags | == == == == == == 81 00 ==
| .——————' '——————.
marker bits | 10-- ---- ---- ----
PES_scrambling_control | --00 ---- ---- ----
PES_priority | ---- 0--- ---- ----
data_alignment_indicator | ---- -0-- ---- ----
copyright | ---- --0- ---- ----
original_or_copy | ---- ---1 ---- ----
PTS_DTS_flags | ---- ---- 00-- ----
ESCR_flag | ---- ---- --0- ---- //elementary
stream clock reference
ES_rate_flag | ---- ---- ---0 ----
DSM_trick_mode_flag | ---- ---- ---- 0---
additional_copy_info_flag | ---- ---- ---- -0--
PES_CRC_flag | ---- ---- ---- --0-
PES_extension_flag | ---- ---- ---- ---0 //1 = pes
extension present
PES_header_data_length | == == == == == == == == 00 //= zero bytes
------------------------------|
Re: H.262 §6.2.3 | PICTURE_HEADER_____________________
| 00 00 01 00 00 9F FF FB B8 00 00 00
picture_start_code | 00 00 01 00 == == == == == == == ==
| == == == == 00 9F FF FB B8 == == ==
| .——————————'
'———————————————————————.
temporal_reference | 0000 0000 10-- ---- ---- ---- ---- ---- ----
----
picture_coding_type | ---- ---- --01 1--- ---- ---- ---- ---- ---- ---- //'001':
I-picture; '010': P-picture; '011': B-picture
vbv_delay | ---- ---- ---- -111 1111 1111 1111 1--- ---- ---- //= no delay;
else, the number of 90 KHz clocks from picture_start_code to decoder start
full_pel_forward_vector | ---- ---- ---- ---- ---- ---- ---- -0-- ----
---- //ignored
forward_f_code | ---- ---- ---- ---- ---- ---- ---- --11 1---
---- //ignored
full_pel_backward_vector | ---- ---- ---- ---- ---- ---- ---- ---- -0--
---- //ignored
backward_f_code | ---- ---- ---- ---- ---- ---- ---- ---- --11
1--- //ignored
extra_bit_picture | ---- ---- ---- ---- ---- ---- ---- ---- ---- -0-- //= no
content_description_data; '1' = followed by 8 bits of content_description_data followed by another
extra_bit_picture
padding_bits | ---- ---- ---- ---- ---- ---- ---- ---- ----
--00
padding bytes | == == == == == == == == == 00 00 00
------------------------------|
Here is a GOP header with 'closed_gop' & 'broken_link'
------------------------------|
Re: H.262 §6.2.2.6 & §6.3.8 | GOP_HEADER_____________
group_start_code | 00 00 01 B8 == == == ==
time_code | == == == == 00 08 00 00 //Re: H.262 §6.3.8
| .—————————' '—————————————————.
drop_frame_flag | 0--- ---- ---- ---- ---- ---- ---- ----
time_code_hours | -000 00-- ---- ---- ---- ---- ---- ---- //= 0
time_code_minutes | ---- --00 0000 ---- ---- ---- ---- ---- //= 0
marker_bit | ---- ---- ---- 1--- ---- ---- ---- ----
time_code_seconds | ---- ---- ---- -000 000- ---- ---- ---- //= 0
time_code_pictures | ---- ---- ---- ---- ---0 0000 0--- ---- //= 0; time code =
00:00:00.00
closed_gop | ---- ---- ---- ---- ---- ---- -0-- ----
broken_link | ---- ---- ---- ---- ---- ---- --0- ---- //'1': GOP has been
edited and B-frames have lost their reference frames
????? | ---- ---- ---- ---- ---- ---- ---0 0000
FFmpeg appears to be making up DTSes & PTSes for P- & B-frames. I can see why it wants to do that,
but the logic of how it makes up DTSes & PTSes is unknown and is suspect. Also, why FFmpeg is not
using the 'closed_gop' & 'broken_link' flags to fix open B-frames is unknown.
--Mark.
_______________________________________________
ffmpeg-user mailing list
ffmpeg-user@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-user
To unsubscribe, visit link above, or email
ffmpeg-user-requ...@ffmpeg.org with subject "unsubscribe".
