Recherche avancée

Sur d’autres sites (10839)

• why is ffmpeg so fast

  8 juillet 2017, par jx.xu

  I have written a ffmpeg-based C++ program about converting yuv to rgb using libswscale, similar to the official example.
  Just simple copy and modify the official example before build in my visual studio 2017 on windows 10. However, the time performance is much slower than the ffmpeg.exe executable file, as 36 seconds vs 12 seconds.
  I already know that ffmpeg uses some optimization techniques like SIMD instructions. While in my performance profiling, the bottleneck is disk I/O writing which takes at least 2/3 time.
  Then I develop a concurrent version where a dedicated thread will handle all I/O task while the situation doesn’t seem to improve. It’s worth noting that I use Boost C++ library to utilize multi-thread and asynchronous events.

  So, I just wanna know how can I modify program using the libraries of ffmpeg or the time performance gap towards ffmpeg.exe just can’ be catched up.

  As requested by friendly answers, I post my codes. Compiler is msvc in vs2017 and I turn on the full optimization /Ox .

  Supplement my main question, I make another plain disk I/O test merely copying the file of the same size. It’s surprising to find that plain sequential disk I/O costs 28 seconds while the front codes cost 36 seconds in total... Any one knows how can ffmpeg finishes the same job in only 12 seconds ? That must use some optimization techniques, like random disk I/O or memory buffer reusing ?

  #include "stdafx.h"
  
  #define __STDC_CONSTANT_MACROS
  
  extern "C" {
  
  #include <libavutil></libavutil>imgutils.h>
  
  #include <libavutil></libavutil>parseutils.h>
  
  #include <libswscale></libswscale>swscale.h>
  
  }
  
  
  
  #ifdef _WIN64
  
  #pragma comment(lib, "avformat.lib")
  
  #pragma comment(lib, "avcodec.lib")
  
  #pragma comment(lib, "avutil.lib")
  
  #pragma comment(lib, "swscale.lib")
  
  #endif
  
  
  
  #include <common></common>cite.hpp>   // just include headers of c++ STL/Boost
  
  
  
  int main(int argc, char **argv)
  
  {
  
      chrono::duration<double> period;
  
      auto pIn = fopen("G:/Panorama/UHD/originalVideos/DrivingInCountry_3840x1920_30fps_8bit_420_erp.yuv", "rb");
  
      auto time_mark = chrono::steady_clock::now();
  
  
  
      int src_w = 3840, src_h = 1920, dst_w, dst_h;
  
      enum AVPixelFormat src_pix_fmt = AV_PIX_FMT_YUV420P, dst_pix_fmt = AV_PIX_FMT_RGB24;
  
      const char *dst_filename = "G:/Panorama/UHD/originalVideos/out.rgb";
  
      const char *dst_size = "3840x1920";
  
      FILE *dst_file;
  
      int dst_bufsize;
  
      struct SwsContext *sws_ctx;
  
      int i, ret;
  
      if (av_parse_video_size(&amp;dst_w, &amp;dst_h, dst_size) &lt; 0) {
  
          fprintf(stderr,
  
              "Invalid size '%s', must be in the form WxH or a valid size abbreviation\n",
  
              dst_size);
  
          exit(1);
  
      }
  
      dst_file = fopen(dst_filename, "wb");
  
      if (!dst_file) {
  
          fprintf(stderr, "Could not open destination file %s\n", dst_filename);
  
          exit(1);
  
      }
  
      /* create scaling context */
  
      sws_ctx = sws_getContext(src_w, src_h, src_pix_fmt,
  
          dst_w, dst_h, dst_pix_fmt,
  
          SWS_BILINEAR, NULL, NULL, NULL);
  
      if (!sws_ctx) {
  
          fprintf(stderr,
  
              "Impossible to create scale context for the conversion "
  
              "fmt:%s s:%dx%d -> fmt:%s s:%dx%d\n",
  
              av_get_pix_fmt_name(src_pix_fmt), src_w, src_h,
  
              av_get_pix_fmt_name(dst_pix_fmt), dst_w, dst_h);
  
          ret = AVERROR(EINVAL);
  
          exit(1);
  
      }
  
      io_service srv;   // Boost.Asio class
  
      auto work = make_shared(srv); 
  
      thread t{ bind(&amp;io_service::run,&amp;srv) }; // I/O worker thread
  
      vector> result;
  
      /* utilize function class so that lambda can capture itself */
  
      function)> recursion;  
  
      recursion = [&amp;](int left, unique_future<bool> writable)   
  
      {
  
          if (left &lt;= 0)
  
          {
  
              writable.wait();
  
              return;
  
          }
  
          uint8_t *src_data[4], *dst_data[4];
  
          int src_linesize[4], dst_linesize[4];
  
          /* promise-future pair used for thread synchronizing so that the file part is written in the correct sequence  */
  
          promise<bool> sync; 
  
          /* allocate source and destination image buffers */
  
          if ((ret = av_image_alloc(src_data, src_linesize,
  
              src_w, src_h, src_pix_fmt, 16)) &lt; 0) {
  
              fprintf(stderr, "Could not allocate source image\n");
  
          }
  
          /* buffer is going to be written to rawvideo file, no alignment */
  
          if ((ret = av_image_alloc(dst_data, dst_linesize,
  
              dst_w, dst_h, dst_pix_fmt, 1)) &lt; 0) {
  
              fprintf(stderr, "Could not allocate destination image\n");
  
          }
  
          dst_bufsize = ret;
  
          fread(src_data[0], src_h*src_w, 1, pIn);
  
          fread(src_data[1], src_h*src_w / 4, 1, pIn);
  
          fread(src_data[2], src_h*src_w / 4, 1, pIn);
  
          result.push_back(async([&amp;] {
  
              /* convert to destination format */
  
              sws_scale(sws_ctx, (const uint8_t * const*)src_data,
  
                  src_linesize, 0, src_h, dst_data, dst_linesize);
  
              if (left>0)
  
              {
  
                  assert(writable.get() == true);
  
                  srv.post([=]
  
                  {
  
                      /* write scaled image to file */
  
                      fwrite(dst_data[0], 1, dst_bufsize, dst_file);
  
                      av_freep((void*)&amp;dst_data[0]);
  
                  });
  
              }
  
              sync.set_value(true);
  
              av_freep(&amp;src_data[0]);
  
          }));
  
          recursion(left - 1, sync.get_future());
  
      };
  
  
  
      promise<bool> root;
  
      root.set_value(true);
  
      recursion(300, root.get_future());  // .yuv file only has 300 frames
  
      wait_for_all(result.begin(), result.end()); // wait for all unique_future to callback
  
      work.reset();    // io_service::work releses
  
      srv.stop();      // io_service stops
  
      t.join();        // I/O thread joins
  
      period = steady_clock::now() - time_mark;  // calculate valid time
  
      fprintf(stderr, "Scaling succeeded. Play the output file with the command:\n"
  
          "ffplay -f rawvideo -pix_fmt %s -video_size %dx%d %s\n",
  
          av_get_pix_fmt_name(dst_pix_fmt), dst_w, dst_h, dst_filename);
  
      cout &lt;&lt; "period" &lt;&lt; et &lt;&lt; period &lt;&lt; en;
  
  end:
  
      fclose(dst_file);
  
      //  av_freep(&amp;src_data[0]);
  
      //  av_freep(&amp;dst_data[0]);
  
      sws_freeContext(sws_ctx);
  
      return ret &lt; 0;
  
  }
  
  </bool></bool></bool></double>

• Transcode to opus by Fluent-ffmpeg or ffmpeg from command line

  14 juillet 2017, par shamaleyte

  My purpose is to transcode a webm file into opus file.
  It works just fine as the following ;

  ffmpeg -i input.webm -vn -c:a copy output.opus

  But the generated opus file always starts from 4rd or 5th seconds when I play it. It seems like that the first seconds are lost. Any idea why it happens ?

  >ffmpeg -i x.webm -vn -c:a copy x1.opus
  
  ffmpeg version N-86175-g64ea4d1 Copyright (c) 2000-2017 the FFmpeg 
  
  developers
  
  built with gcc 6.3.0 (GCC)
  
  configuration: --enable-gpl --enable-version3 --enable-cuda --enable-cuvid -
  
  -enable-d3d11va --enable-dxva2 --enable-libmfx --enable-nvenc --enable-
  
  avisynth --enable-bzlib --enable-fontconfig --enable-frei0r --enable-gnutls 
  
  --enable-iconv --enable-libass --enable-libbluray --enable-libbs2b --enable-
  
  libcaca --enable-libfreetype --enable-libgme --enable-libgsm --enable-
  
  libilbc --enable-libmodplug --enable-libmp3lame --enable-libopencore-amrnb -
  
  -enable-libopencore-amrwb --enable-libopenh264 --enable-libopenjpeg --
  
  enable-libopus --enable-librtmp --enable-libsnappy --enable-libsoxr --
  
  enable-libspeex --enable-libtheora --enable-libtwolame --enable-libvidstab -
  
  -enable-libvo-amrwbenc --enable-libvorbis --enable-libvpx --enable-
  
  libwavpack --enable-libwebp --enable-libx264 --enable-libx265 --enable-
  
  libxavs --enable-libxvid --enable-libzimg --enable-lzma --enable-zlib
  
    libavutil      55. 63.100 / 55. 63.100
  
    libavcodec     57. 96.101 / 57. 96.101
  
    libavformat    57. 72.101 / 57. 72.101
  
    libavdevice    57.  7.100 / 57.  7.100
  
    libavfilter     6. 90.100 /  6. 90.100
  
    libswscale      4.  7.101 /  4.  7.101
  
    libswresample   2.  8.100 /  2.  8.100
  
    libpostproc    54.  6.100 / 54.  6.100
  
  Input #0, matroska,webm, from 'x.webm':
  
    Metadata:
  
    encoder         : libwebm-0.2.1.0
  
    creation_time   : 2017-06-19T20:50:21.722000Z
  
  Duration: 00:00:32.33, start: 0.000000, bitrate: 134 kb/s
  
  Stream #0:0(eng): Audio: opus, 48000 Hz, mono, fltp (default)
  
  Stream #0:1(eng): Video: vp8, yuv420p(progressive), 640x480, SAR 1:1 DAR 
  
  4:3, 16.67 fps, 16.67 tbr, 1k tbn, 1k tbc (default)
  
  Output #0, opus, to 'x1.opus':
  
  Metadata:
  
  encoder         : Lavf57.72.101
  
  Stream #0:0(eng): Audio: opus, 48000 Hz, mono, fltp (default)
  
  Metadata:
  
    encoder         : Lavf57.72.101
  
  Stream mapping:
  
  Stream #0:0 -> #0:0 (copy)
  
  Press [q] to stop, [?] for help
  
  size=     114kB time=00:00:32.33 bitrate=  28.8kbits/s speed=3.22e+003x
  
  video:0kB audio:111kB subtitle:0kB other streams:0kB global headers:0kB 
  
  muxing overhead: 2.152229%

  It is jumping from 0 to 4th second .
  Please take a look at this screencast.
  https://www.screenmailer.com/v/52IXnpAarHavwJE

  This is the sample video file that I tried to transcode : https://drive.google.com/open?id=0B2sa3oV_Y3X_ZmVWX3MzTlRPSmc

  So I guess the transcoding starts right at the point that the voice comes in, why is that ?

• Video encoder & segmenter for HLS VoD poor quality

  10 juillet 2017, par Murilo

  I am trying to encode and segment video for HLS on demand(VoD).
  I am using the following code for such :

  ffmpeg -i 20170706_174314.mp4 -c 24 \
  
          -vcodec libx264 -acodec aac -ac 1 -strict -2 -b:v 128k \
  
          -profile:v baseline -maxrate 400k -bufsize 1835k \
  
          -hls_time 10 -hls_playlist_type vod -vsync 1 \
  
          video_chunks/index1.m3u8 \
  
          -c 24 -vcodec libx264 -acodec aac -ac 1 -strict -2 -b:v 128k \
  
          -profile:v baseline -maxrate 700k -bufsize 1835k \
  
          -hls_time 10 -hls_playlist_type vod -vsync 1 \
  
          video_chunks/index2.m3u8

  I tried this other code also just for segmenting but had the same exactly problem :

  ffmpeg -i 20170706_174314.mp4 \
  
  -c:a libmp3lame -ar 48000 -ab 64k  -c:v libx264 -b:v 128k -flags \ 
  
  -global_header -map 0 -f segment \
  
  -segment_list video_chunks/test.m3u8 -segment_time 10 -segment_format mpegts \
  
  video_chunks/segment_%05d.ts

  Later on I create another playlist with bandwidth separators to call on the two other playlists generated with the code above.

  This code was working great on some videos but yesterday I recorded a video with my Samsung J7 Prime phone to test since the videos will be generated by phone and this video was poorly encoded. The quality sucks and some parts of the video turned Black&White.

  Another thing I noticed on this video is that the following message kept appearing in loop until the end of the encoding&segmenting process.

  Past duration X too large

  Where X is a decimal really close to

  0.675316

  The link to the video is below :

  Dropbox Link

  My FFmpeg version :

  ffmpeg --version
  
  ffmpeg version N-86482-gbc40674 Copyright (c) 2000-2017 the FFmpeg developers
  
    built with gcc 7.1.0 (GCC)
  
    configuration: --enable-gpl --enable-version3 --enable-cuda --enable-cuvid --enable-d3d11va --enable-dxva2 --enable-libmfx --enable-nvenc --enable-avisynth --enable-bzlib --enable-fontconfig --enable-frei0r --enable-gnutls --enable-iconv --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --enable-libfreetype --enable-libgme --enable-libgsm --enable-libilbc --enable-libmodplug --enable-libmp3lame --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenh264 --enable-libopenjpeg --enable-libopus --enable-librtmp --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libtheora --enable-libtwolame --enable-libvidstab --enable-libvo-amrwbenc --enable-libvorbis --enable-libvpx --enable-libwavpack --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxavs --enable-libxvid --enable-libzimg --enable-lzma --enable-zlib
  
    libavutil      55. 66.100 / 55. 66.100
  
    libavcodec     57. 99.100 / 57. 99.100
  
    libavformat    57. 73.100 / 57. 73.100
  
    libavdevice    57.  7.100 / 57.  7.100
  
    libavfilter     6. 92.100 /  6. 92.100
  
    libswscale      4.  7.101 /  4.  7.101
  
    libswresample   2.  8.100 /  2.  8.100
  
    libpostproc    54.  6.100 / 54.  6.100

  SO : Windows 10

  EDIT1 : Link to the output
  If you see the output it might be worth saying I am also seeing the message

  VBV underflow(Frame X, -Y bits)