Recherche avancée

Sur d’autres sites (9283)

• lavc/aarch64 : port HEVC SIMD idct NEON

  16 janvier 2021, par Reimar Döffinger

  lavc/aarch64 : port HEVC SIMD idct NEON
  Makes SIMD-optimized 8x8 and 16x16 idcts for 8 and 10 bit depth
  
  available on aarch64.
  
  For a UHD HDR (10 bit) sample video these were consuming the most time
  
  and this optimization reduced overall decode time from 19.4s to 16.4s,
  
  approximately 15% speedup.
  
  Test sample was the first 300 frames of "LG 4K HDR Demo - New York.ts",
  
  running on Apple M1.
  Signed-off-by : Josh Dekker <josh@itanimul.li>
  

  • [DH] libavcodec/aarch64/Makefile
  • [DH] libavcodec/aarch64/hevcdsp_idct_neon.S
  • [DH] libavcodec/aarch64/hevcdsp_init_aarch64.c
  • [DH] libavcodec/hevcdsp.c
  • [DH] libavcodec/hevcdsp.h

• ffmpeg -ss then apply filter then concat producing timestamp errors

  6 août 2020, par Bob Ramsey

  Using ffmpeg, I have split a file into multiple parts using -ss. Then I apply a filter to some of the files, then concat the files back together. When I do that, I get : Non-monotonous DTS in output stream 0:0 ; previous : 341334, current : 340526 ; changing to 341335. This may result in incorrect timestamps in the output file. The output file plays, but there are noticeable skips where the files are joined.

  


  Here's how I am splitting the file :

  


  ffmpeg -i full_source.mp4 -ss 0 -to 14.264250 -c copy  01-plain.mp4
ffmpeg -i full_source.mp4 -ss 14.264250 -to 18.435083 -c copy  01-filtered.mp4

ffmpeg -i full_source.mp4 -ss 18.435083 -to 29.988292 -c copy  02-plain.mp4
ffmpeg -i full_source.mp4 -ss 29.988292 -to 31.865167 -c copy  02-filtered.mp4
...
ffmpeg -i full_source.mp4 -ss 0 -to 14.264250 -c copy  10-plain.mp4
ffmpeg -i full_source.mp4 -ss 234.484203 -to 300.000 -c copy  10-filtered.mp4


  


  Then I apply a different drawtext filter on each of the 10 filtered files and save them with a new name, like :

  


  ffmpeg -hide_banner -loglevel warning -y -i 01-filtered.mp4 -filter_complex "drawtext=fontfile=calibri.ttf:fontsize=24:fontcolor=white:x=300:y=500:text='hello world'" -crf 15 01-filtered-complete.mp4


  


  Finally, I join all of the plain and complete files back together like this :

  


  ffmpeg  -f concat -safe 0 -i mylist.txt -c:a copy -c:v copy  outfile.mp4


  


  And that's where the timing error comes in. I've tried adding -vsync drop in the concat command, but that didn't really work either. Same version of ffmpeg does the split, the filter, and the concat. I've tried different versions, everything from 20170519 to one from May 2020 with the same result. Always making sure that all three steps are done by the same version of ffmpeg.

  


  The only thing I can see is that ffprobe shows a duration of 14.27 for 01-plain.mp4 when it should be 14.264250. All of the other files show a similar rounding difference. The files are 23.98 fps. If I do all of my filters in really long command without splitting the file, I can use the more precise numbers with no problem. It just takes 10 times as long. This is all scripted, it happens a couple of hundred times a day and time is money, so I can't take 10 times as long to do each file.

  


  Any ideas ? Thanks in advance !

  


• hlsenc : Calculate the average and actual maximum bitrate of segments

  20 juin 2024, par Martin Storsjö

  hlsenc : Calculate the average and actual maximum bitrate of segments
  Previously, the bitrate advertised in the master playlist would only
  
  be based on the nominal values in either AVCodecParameters bit_rate,
  
  or via AVCPBProperties max_bitrate. On top of this, a
  
  fudge factor of 10% is added, to account for container overhead.
  Neither of these bitrates may be known, and if the encoder is
  
  running in VBR mode, there is no such value to be known. And
  
  the container overhead may be more or less than the given
  
  constant factor of 10%.
  Instead, calculate the maximum bitrate per segment based on
  
  what actually gets output from the muxer, and average bitrate
  
  across all segments.
  When muxing of the file finishes, update the master playlist
  
  with these values, exposing both the maximum (which previously
  
  was a guesstimate based on the nominal values) via
  
  EXT-X-STREAM-INF BANDWIDTH, and the average via
  
  EXT-X-STREAM-INF AVERAGE-BANDWIDTH.
  This makes it possible to use the hlsenc muxer with VBR
  
  encodes, for VOD style muxing.
  Signed-off-by : Martin Storsjö <martin@martin.st>
  

  • [DH] libavformat/dashenc.c
  • [DH] libavformat/hlsenc.c
  • [DH] libavformat/hlsplaylist.c
  • [DH] libavformat/hlsplaylist.h