Public/removeads
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

H264 concatenation functions are moved to cutting module.

Browse Source
This commit is contained in:
Frédéric Tronel
2026-08-30 17:45:03 +02:00
parent 34d1bc26f3
commit d756fede54
5 changed files with 93 additions and 122 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/tscut/cli.py
+1 -1
View File
@@ -12,8 +12,8 @@ from pathlib import Path
import coloredlogs
from tscut.models import ProcessingOptions
from tscut.exceptions import TSCutError
from tscut.models import ProcessingOptions
from tscut.pipeline import process_recording
from tscut.tools.discovery import check_required_tools
from tscut.tools.timeframe import compare_time_interval, parse_time_interval
src/tscut/coarse.py
-6
View File
@@ -12,16 +12,10 @@ from tscut.tools.mkvtoolnix import (
logger = logging.getLogger(__name__)
# TODO: finish this procedure
def do_coarse_processing(ffmpeg_path:str, ffprobe_path:str, mkvmerge_path:str,
input_file: IO[bytes], begin, end, nb_frames, framerate,
files_prefix, streams, width, height, temporaries, dump_mem_fd) -> None:
# pylint: disable=W0613
logger = logging.getLogger(__name__)
# Internal video with all streams (video, audio and subtitles)
internal_mkv_name = f'{files_prefix}.mkv'
src/tscut/cutting.py
+66 -2
View File
@@ -4,10 +4,19 @@
import logging
from datetime import timedelta
from os import (
SEEK_SET,
fstat,
lseek,
read,
write,
)
from shutil import copyfile
from typing import IO, Any
from typing import IO, Any, BinaryIO, Sequence, TextIO
import hexdump
from tqdm import tqdm
from typeguard import typechecked
from tscut.exceptions import InvalidMediaError
from tscut.h264.avc import get_avc_config_from_h264
@@ -25,11 +34,66 @@ from tscut.tools.mkvtoolnix import (
remove_video_tracks_from_mkv,
)
from tscut.tools.timeframe import get_ts_frame
from tscut.tscut import concatenate_h264_parts, concatenate_h264_ts_parts
logger = logging.getLogger(__name__)
@typechecked
def concatenate_h264_parts(h264parts: Sequence[BinaryIO], output: BinaryIO) -> None:
logger = logging.getLogger(__name__)
total_length = 0
for h264 in h264parts:
fd = h264.fileno()
total_length += fstat(fd).st_size
logger.info('Total length: %d', total_length)
outfd = output.fileno()
lseek(outfd, 0, SEEK_SET)
pb = tqdm(total=total_length, unit='bytes', desc='Concatenation')
for h264 in h264parts:
fd = h264.fileno()
lseek(fd, 0, SEEK_SET)
while True:
buf = read(fd, 1000000)
if buf is None or len(buf) == 0:
break
pos = 0
while pos < len(buf):
nb_bytes = write(outfd, buf[pos:])
pb.update(nb_bytes)
pos += nb_bytes
def concatenate_h264_ts_parts(h264_ts_parts: Sequence[TextIO], output: TextIO) -> None:
logger = logging.getLogger(__name__)
header = '# timestamp format v2\n'
output.write(header)
last = 0.
first = True
for part in h264_ts_parts:
if first:
offset = last
else:
# TODO: take framerate into account
offset = last + 40
logger.debug('Parsing file: %s. Offset=%d', part, offset)
isheader = part.readline()
if (not isheader) or (isheader != header):
logger.error('Impossible to find a valid header: "%s"', isheader)
exit(-1)
while True:
line = part.readline()
if not line:
break
ts = offset + float(line)
last = max(last,ts)
output.write(f'{ts:f}\n')
if first:
first = False
def cut_recording(media: PreparedMedia, options: ProcessingOptions,
temporaries: list[IO[Any]]) -> CutResult:
src/tscut/docs/README.md
+26
View File
@@ -0,0 +1,26 @@
Useful SPS/PPS discussion:
1.https://copyprogramming.com/howto/including-sps-and-pps-in-a-raw-h264-track
2. https://gitlab.com/mbunkus/mkvtoolnix/-/issues/2390
Strategy: a possible way of handling multiple SPS/PPS gracefully.
Encode each head and trailer with FFMPEG using only I-frame (to be sure the NAL unit will never refer to another image).
Encode using an different SPS-ID all of them (using sps-id parameter of libx264 library, e.g 1 instead of 0).
For the video track produce only a raw H264 file and a file containing timestamps of the different frames.
For the rest of the tracks (audio, subtitles) produce directly a MKV (this is already done).
Concatenate all raw H264 in a giant one (like cat), and the same for timestamps of video frames (to keep sound and video synchronized).
Then use mkvmerge to remux the H264 track and the rest of tracks.
MKVmerge "concatenate" subcommand is able to concatenate different SPS/PPS data into a bigger Private Codec Data.
However, this is proved to be not reliable.
Sometimes it results in a AVC context containing a single SPS/PPS.
So we have to rely on a manual parsing of the H264 AVC context of original movie and the ones produced for headers and trailers, and then merging them into a bigger AVC context.
Then finally, change the Private Codec Data in the final MKV.
Extract SPS/PPS:
1. https://gitlab.com/mbunkus/mkvtoolnix/-/issues/2390
```bash
ffmpeg -i <InputFile (before concatenation)> -c:v copy -an -sn -bsf:v trace_headers -t 0.01 -report -loglevel 0 -f null -
``
src/tscut/tscut.py
-113
View File
@@ -1,113 +0,0 @@
#!/usr/bin/env python3
'''A module to remove parts of video (.e.g advertisements) with single frame precision.'''
# Standard modules
import logging
from os import (
SEEK_SET,
fstat,
lseek,
read,
write,
)
from sys import exit
from typing import IO, BinaryIO, Sequence, TextIO
# Third party libraries
from tqdm import tqdm
from typeguard import typechecked
from tscut.models import SupportedFormat
from tscut.tools.mkvtoolnix import (
extract_mkv_part,
)
# Useful SPS/PPS discussion.
# https://copyprogramming.com/howto/including-sps-and-pps-in-a-raw-h264-track
# https://gitlab.com/mbunkus/mkvtoolnix/-/issues/2390
# New strategy: a possible way of handling multiple SPS/PPS gracefully.
# Encode each head and trailer with FFMPEG using only I-frame (to be sure the NAL unit will never
# refer to another image).
# Encode using an different SPS-ID all of them (using sps-id parameter of libx264 library, e.g
# 1 instead of 0).
# For the video track produce only a raw H264 file and a file containing timestamps of the
# different frames.
# For the rest of the tracks (audio, subtitles) produce directly a MKV (this is already done).
# Concatenate all raw H264 in a giant one (like cat), and the same for timestamps of video frames
# (to keep sound and video synchronized).
# Then use mkvmerge to remux the H264 track and the rest of tracks.
# MKVmerge "concatenate" subcommand is able to concatenate different SPS/PPS data into a bigger
# Private Codec Data.
# However, this is proved to be not reliable. Sometimes it results in a AVC context containing
# a single SPS/PPS.
# So we have to rely on a manual parsing of the H264 AVC context of original movie
# and the ones produced for headers and trailers, and then merging them into a bigger AVC context.
# Then finally, change the Private Codec Data in the final MKV.
# Extract SPS/PPS
# https://gitlab.com/mbunkus/mkvtoolnix/-/issues/2390
# ffmpeg -i <InputFile (before concatenation)> -c:v copy -an -sn -bsf:v trace_headers -t 0.01\
# -report -loglevel 0 -f null -
@typechecked
def concatenate_h264_parts(h264parts: Sequence[BinaryIO], output: BinaryIO) -> None:
logger = logging.getLogger(__name__)
total_length = 0
for h264 in h264parts:
fd = h264.fileno()
total_length += fstat(fd).st_size
logger.info('Total length: %d', total_length)
outfd = output.fileno()
lseek(outfd, 0, SEEK_SET)
pb = tqdm(total=total_length, unit='bytes', desc='Concatenation')
for h264 in h264parts:
fd = h264.fileno()
lseek(fd, 0, SEEK_SET)
while True:
buf = read(fd, 1000000)
if buf is None or len(buf) == 0:
break
pos = 0
while pos < len(buf):
nb_bytes = write(outfd, buf[pos:])
pb.update(nb_bytes)
pos += nb_bytes
def concatenate_h264_ts_parts(h264_ts_parts: Sequence[TextIO], output: TextIO) -> None:
logger = logging.getLogger(__name__)
header = '# timestamp format v2\n'
output.write(header)
last = 0.
first = True
for part in h264_ts_parts:
if first:
offset = last
else:
# TODO: take framerate into account
offset = last + 40
logger.debug('Parsing file: %s. Offset=%d', part, offset)
isheader = part.readline()
if (not isheader) or (isheader != header):
logger.error('Impossible to find a valid header: "%s"', isheader)
exit(-1)
while True:
line = part.readline()
if not line:
break
ts = offset + float(line)
last = max(last,ts)
output.write(f'{ts:f}\n')
if first:
first = False