Recherche avancée

Sur d’autres sites (8379)

• Merge multiple mp4 (h264 + aac) videos into one frame without re-encoding

  1er décembre 2017, par shubhadeep banerjee

  I have four mp4 files with same length (1 minute each) , say 1.mp4, 2.mp4, 3.mp4 and 4.mp4. I want to merge these four videos into a single frame and create a mosaic of four videos ; 1.mp4 at the top-left, 2.mp4 at the top-right, 3.mp4 at the bottom-left and 4.mp4 at the bottom-right.

  I have tried the same using ffmpeg filter_complex.

  ffmpeg -i 1.mp4 -i 2.mp4 -i 3.mp4 -i 4.mp4 -filter_complex "
  nullsrc=size=640x480 [base] ; [0:v] setpts=PTS-STARTPTS,
  scale=320x240 [upperleft] ; [1:v] setpts=PTS-STARTPTS, scale=320x240
  [upperright] ; [2:v] setpts=PTS-STARTPTS, scale=320x240 [lowerleft] ;
  [3:v] setpts=PTS-STARTPTS, scale=320x240 [lowerright] ;
  [base][upperleft] overlay=shortest=1 [tmp1] ; [tmp1][upperright]
  overlay=shortest=1:x=320 [tmp2] ; [tmp2][lowerleft]
  overlay=shortest=1:y=240 [tmp3] ; [tmp3][lowerright]
  overlay=shortest=1:x=320:y=240 " output.mp4

  But the problem is, ffmpeg re-encodes the video. The input videos already have h.264 + aac encoding, and the output video will have the same encoding. I would be running this code on a very low-power embedded Linux board which have only one core CPU. The board is taking almost 6x times of the file length to produce the output (i.e. 6 minutes in my case), and also consuming 100% of the CPU.

  Is there any other way to do this, without re-encoding ? I run the whole thing on Debian, so I can use any package you suggest.

• Repair mpeg files using ffmpeg

  27 février 2014, par rsdrsd

  I have a bunch of mpeg files which are some how invalid or incorrect. I can play the files in different media players but when I upload the files and they should automagically be converted. It takes a very long time to create screenshots and it creates about 10000 screenshots instead of the 50 to be expected. The command is part of an automatic conversion app. With mp4 and other files it works great but whit mpeg it doesn't work as expected. The creation of screenshots eats up all memory and processor power.

  For creating screenshots I have tried the following :

      ffmpeg -y -i /input/file.mpeg -f image2 -aspect 16:9 -bt 20M -vsync passthrough -vf select='isnan(prev_selected_t)+gte(t-prev_selected_t\,10)' /output/file-%05d.jpg

  this just creates 2 screenshots while I expect 50 or so. The following command :

      ffmpeg -y -i /input/file.mpeg -f image2 -vf fps=fps=1/10 -aspect 16:9 -vsync passthrough -bt 20M /output/file-%05d.jpg

  gave me errors about buffers :

      ffmpeg version N-39361-g1524b0f Copyright (c) 2000-2014 the FFmpeg developers
  
        built on Feb 26 2014 23:46:40 with gcc 4.4.7 (GCC) 20120313 (Red Hat 4.4.7-4)
  
        configuration: --prefix=/home/example/ffmpeg_build --extra-cflags=-I/home/example/ffmpeg_build/include --extra-ldflags=-L/home/example/ffmpeg_build/lib --bindir=/home/example/bin --extra-libs=-ldl --enable-gpl --enable-nonfree --enable-libfdk_aac --enable-libmp3lame --enable-libopus --enable-libvorbis --enable-libvpx --enable-libx264 --enable-libfreetype --enable-libspeex --enable-libtheora
  
        libavutil      52. 66.100 / 52. 66.100
  
        libavcodec     55. 52.102 / 55. 52.102
  
        libavformat    55. 33.100 / 55. 33.100
  
        libavdevice    55. 10.100 / 55. 10.100
  
        libavfilter     4.  2.100 /  4.  2.100
  
        libswscale      2.  5.101 /  2.  5.101
  
        libswresample   0. 18.100 /  0. 18.100
  
        libpostproc    52.  3.100 / 52.  3.100
  
      [mp3 @ 0x200d7c0] Header missing
  
      [mpegts @ 0x2008a60] DTS discontinuity in stream 0: packet 6 with DTS 34185, packet 7 with DTS 8589926735
  
      [mpegts @ 0x2008a60] Invalid timestamps stream=0, pts=7157, dts=8589932741, size=150851
  
      Input #0, mpegts, from '/home/example/app/uploads/21.mpeg':
  
        Duration: 00:03:14.75, start: 0.213000, bitrate: 26112 kb/s
  
        Program 1
  
  Stream #0:0[0x3e9]: Video: mpeg2video (Main) ([2][0][0][0] / 0x0002), yuv420p(tv), 1440x1080 [SAR 4:3 DAR 16:9], max. 25000 kb/s, 29.97 fps, 60 tbr, 90k tbn, 59.94 tbc
  
  Stream #0:1[0x3ea]: Audio: mp2 ([3][0][0][0] / 0x0003), 48000 Hz, stereo, s16p, 384 kb/s
  
      [swscaler @ 0x1ff9860] deprecated pixel format used, make sure you did set range correctly
  
      Output #0, image2, to '/home/example/app/uploads/21-%05d.jpg':
  
        Metadata:
  
  encoder         : Lavf55.33.100
  
  Stream #0:0: Video: mjpeg, yuvj420p, 1440x1080 [SAR 4:3 DAR 16:9], q=2-31, 200 kb/s, 90k tbn, 0.10 tbc
  
      Stream mapping:
  
        Stream #0:0 -> #0:0 (mpeg2video -> mjpeg)
  
      Press [q] to stop, [?] for help
  
      [mpegts @ 0x2008a60] Invalid timestamps stream=0, pts=7157, dts=8589932741, size=150851
  
      [output stream 0:0 @ 0x1ff2ba0] 100 buffers queued in output stream 0:0, something may be wrong.
  
      [output stream 0:0 @ 0x1ff2ba0] 1000 buffers queued in output stream 0:0, something may be wrong.

  and it creates about 10000 screenshots while I expect 50.

  Now I have read some where on how to repair some broken files. For this I have the following command :

      ffmpeg -y -i input.mpeg -codec:v copy -codec:a copy output.mpeg

  This however creates a file somewhat smaller, but if I run the same command on the output again, I would expect that it creates the same file, but the following command

      ffmpeg -y -i output.mpeg -codec:v copy -codec:a copy output2.mpeg

  returns a file which is much smaller and runs for only a few seconds while the original was about 3 minutes.

  If I run the "repair" commands for a not broken mpeg it results the first time I run the command in a much smaller file. With ffprobe I checked what changed but the only thing that changes is MPEG-TS to MPEG-PS.

  If I run the command over an mp4 file it results in exactly the same file as expected. Does someone have a clue of what is going wrong. It is boggling me now for about two days and I really have no clue. Or does someone have a good suggestion on how to extract screenshots every 10 seconds without creating too much screenshots and eating up all memory and processor power.

• lavu/tx : invert permutation lookups

  27 février 2021, par Lynne

  lavu/tx : invert permutation lookups
  out[lut[i]] = in[i] lookups were 4.04 times(!) slower than
  
  out[i] = in[lut[i]] lookups for an out-of-place FFT of length 4096.
  The permutes remain unchanged for anything but out-of-place monolithic
  
  FFT, as those benefit quite a lot from the current order (it means
  
  there's only 1 lookup necessary to add to an offset, rather than
  
  a full gather).
  The code was based around non-power-of-two FFTs, so this wasn't
  
  benchmarked early on.
  

  • [DH] libavutil/tx.c
  • [DH] libavutil/tx_priv.h
  • [DH] libavutil/tx_template.c