Recherche avancée

Sur d’autres sites (11350)

• Filesize difference between ffmpeg command line and C API

  31 mars 2021, par Vuwox

  While using following ffmpeg command :

  


  ffmpeg -framerate 12 -pattern_type glob -i 'FRAMES/frame_*.tif' -map 0:v test_binffmpeg.mp4


  


  It uses the default H.264 parameters which are preset=medium and crf=23. This generates a filesize with bit_rate = 1358 kb/s and size = 1659kb.

  


  But by using my own code to encode an mp4 video (using the exact same images), I then have a video around bit_rate = 3929 kb/s and size = 4757kb.

  


  In my code, I leave the codec options to default, and even if I try to overwrite with preset=medium + crf=23, I still have the same size of 4757kb. Im wondering what am I missing ?

  


  My muxer initialization looks like the following :

  


      // Search for the output format based on the file name.
    AVOutputFormat* oformat = av_guess_format(nullptr, m_filename.c_str(), nullptr);
    if ( !oformat ) return false;

    // Allocate the context based on that format
    _ret = avformat_alloc_output_context2(&m_ofmt_ctx, oformat, nullptr, m_filename.c_str());
    if ( _ret ) return false;

    // Get the codec appropriate to that format
    // and be able to set some parameter to that codec.
    AVCodec* codec = avcodec_find_encoder(oformat->video_codec);
    if ( !codec ) return false;

#if _DEBUG
    printf("Codec name: %s\n", avcodec_get_name(codec->id));
    const AVPixelFormat* p = codec->pix_fmts;
    while ( p != NULL && *p != AV_PIX_FMT_NONE )
    {
        printf("supported pix fmt: %s\n", av_get_pix_fmt_name(*p));
        ++p;
    }
#endif

    // Prepare stream/context according to above variable (codec/format).
    AVStream* stream = avformat_new_stream(m_ofmt_ctx, codec);
    if ( !stream ) return false;

    // Allocate the context for that codec.
    m_o_codec_ctx = avcodec_alloc_context3(codec);
    if ( !m_o_codec_ctx ) return false;        

    // Setup the 'global' parameter for the codec (Size, color format, etc)
    stream->codecpar->codec_id = codec->id;
    stream->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
    stream->codecpar->width = m_width;
    stream->codecpar->height = m_height;
    stream->codecpar->format = AV_PIX_FMT_YUV444P;

    // Assign the global parameter to the context.
    avcodec_parameters_to_context(m_o_codec_ctx, stream->codecpar);

    // Set the framerate.
    m_o_codec_ctx->time_base = { 1, m_framerate };

#if 0 // We can enable this, but it still the default parameters anyway.
    if ( (_ret = av_dict_set(&m_options_dict, "preset", "medium", 0)) < 0 )
        return false;

    if ( (_ret = av_dict_set(&m_options_dict, "crf", "23.0", 0)) < 0 )
        return false;
#endif
    
    avcodec_parameters_from_context(stream->codecpar, m_o_codec_ctx);

    // Open the file and write the header.
    if ( (_ret = avcodec_open2(m_o_codec_ctx, codec, &m_options_dict)) < 0 )
        return false;

    if ( (_ret = avio_open(&m_ofmt_ctx->pb, m_filename.c_str(), AVIO_FLAG_WRITE)) < 0 )
        return false;

    if ( (_ret = avformat_write_header(m_ofmt_ctx, &options_dict)) < 0 )
        return false;

#if _DEBUG
    // This print the stream/output format.
    av_dump_format(m_ofmt_ctx, 0, m_filename.c_str(), 1);
#endif

    // Define the pts increment (the time to the next frame).
    if ( m_ofmt_ctx && m_ofmt_ctx->nb_streams > 0 )
        m_pts_increment = av_rescale_q(1, m_o_codec_ctx->time_base, m_ofmt_ctx->streams[0]->time_base);
    else
       return false;
       
   return true;


  


  EDIT :

  


  As requested user llogan in comment, I provide the ffmpeg -i test_binffmpeg.mp4 -i test_apiffmpeg.mp4 output :

  


  Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'test_binffmpeg.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41
    encoder         : Lavf58.77.100
  Duration: 00:00:10.00, start: 0.000000, bitrate: 1358 kb/s
    Stream #0:0(und): Video: h264 (High 4:4:4 Predictive) (avc1 / 0x31637661), yuv444p, 1080x1080 [SAR 1:1 DAR 1:1], 1356 kb/s, 12 fps, 12 tbr, 12288 tbn, 24 tbc (default)
    Metadata:
      handler_name    : VideoHandler
Input #1, mov,mp4,m4a,3gp,3g2,mj2, from 'test_apiffmpeg.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41
    encoder         : Lavf58.77.100
  Duration: 00:00:09.92, start: 0.000000, bitrate: 3929 kb/s
    Stream #1:0(und): Video: h264 (High 4:4:4 Predictive) (avc1 / 0x31637661), yuv444p, 1080x1080, 3927 kb/s, 12.10 fps, 12 tbr, 90k tbn, 24 tbc (default)


  


  EDIT2 :

  


  I noticed that while using ffprobe over both video I have the following difference :

  


  Params=Bin value ; API value
codec_time_base=1/24;119/2880
avg_frame_rate=12/1;1440/119


  


  Why I have this 119 ? I know that I have 120 frames, and the ffprobe is telling me both file have nb_frames=120.

  


  EDIT3

  


  By modifying the following code :

  


  // Predefine the frame-rate to enforce that usage.
stream->time_base = AVRational{ 1, m_framerate };
stream->r_frame_rate = AVRational{ m_framerate ,1 };
stream->avg_frame_rate = AVRational{ m_framerate ,1 };

// Assign the global parameter to the context.
avcodec_parameters_to_context(m_o_codec_ctx, stream->codecpar);

// Set the framerate.
m_o_codec_ctx->time_base = AVRational{ 1, m_framerate };
m_o_codec_ctx->framerate = AVRational{ m_framerate,  1 };


  


  I was able to enforce the framerate and the above ffprobe informations are now correct, but still have problem with bitrate/filesize.

  


  After investigation, and looking into the logs, I saw that for command line, I have AVG QP around 16-22, and by using the API, I have AVG QP 7-11, which leads to big difference in quantization, and giving for sure a difference in bitrate.

  


  I tried to set -qmin -qmax in my API, and I was able to get something nearby there result, but still different. How can I explain that ?

  


• Difference in number of channels returned by mediainfo and ffprob

  16 mars 2021, par Hai Tran

  I was examining an audio file and noticed that the numbers of channels returned by mediainfo and ffprobe were different.

  


  The mediainfo command :

  


  mediainfo audio.mp4


  


  The ffprobe command (see the channels value) :

  


  ffprobe -i audio.mp4 -show_streams


  


  Does anyone know what's happening ?
Here is the audio file for your own test.

  


• Revision 6fb418741f : Inline xform_quant() in encode_block_intra(). Also inline some of the block cal

  11 juillet 2013, par Ronald S. Bultje

  Changed Paths :
   Modify /vp9/encoder/vp9_encodemb.c

  
  
  Inline xform_quant() in encode_block_intra().

  Also inline some of the block calculations to assist the compiler to
  not do silly things like calculating the same offset (or converting
  between raster/transform block offset or block, mi and pixel unit)
  many, many, many times.

  Cycle times :
  4x4 : 584 -> 505 cycles (16% faster)
  8x8 : 1651 -> 1560 cycles (6% faster)
  16x16 : 7897 -> 7704 cycles (2.5% faster)
  32x32 : 16096 -> 15852 cycles (1.5% faster)

  Overall, this saves about 0.5 seconds (1min49.8 -> 1min49.3) on the
  first 50 frames of bus (speed 0) @ 1500kbps, i.e. 0.5% overall.

  Change-Id : If3dd62453f8e2ab9d4ee616bc4ea956fb8874b80