Recherche avancée

Sur d’autres sites (11066)

• Why multithead use more cpu than single for the same task of deocoing H.264 video ?

  29 juillet 2018, par gary

  I am developing an PC hardware accelerated video decoder for decoding real time H264 Annex B stream. After a bit research, I have done this job by ffmpeg. However, I try to decople the decoding process and show process, the cpu bumped.

  What I want to do is to acceleratly decode a living H.264 stream to yuv, and convert YUV to BGR for cv::Mat without display.

  I can acceleratly decode the h.264 by ffmpeg with decoder h264_qsv, the cpu usage decrease 15% indeed. Then I try to do another thing that put then CV::Mat to std::queue and read the Mat from another queue which just read with do anything else. But the cpu bumped to 30%.

  At first, I think the std::mutex accessed frequently made the performance worse, I try to use a bool flag as lock, but it seems does not work.

  How can I approach ? I have no idear why mulithread make the cpu usage increase. Is there any helpful resources to get clear intuition ?

  frame enqueue

     av_frame_q_lock.lock();
  
     if (av_frame_q.size() == MAX_QUEUE_SIZE) {
  
          av_frame_q.pop();
  
      }
  
      av_frame_q.push(cvframe);
  
      av_frame_q_lock.unlock();

  frame dequeue

      hwrtsp_open("rtsp://admin:123@10.20.37.185/cam/realmonitor?channel=1&subtype=0", "h264_qsv");
  
  cv::Mat frame;
  
  while (true) {
  
      int ret = hwrtsp_read(frame);
  
      if (ret == 0) {
  
      }
  
      cv::waitKey(40);
  
  }

  Thanks in advance !

• How to fill an AVFrame with audio data

  8 mai 2020, par Denis Gottardello

  this is a function taken from an ffmpeg example
How can I fill the frame with my data buffer (const uchar* data, int Length) ?

  



  AVFrame *frame = OutputStreamAudio.pAVFrameTemp;
int j, i, v;
int16_t *q = (int16_t*)frame->data[0];

/* check if we want to generate more frames */
if (av_compare_ts(OutputStreamAudio.next_pts, OutputStreamAudio.pAVCodecContext->time_base, STREAM_DURATION, (AVRational){ 1, 1 }) >= 0) return NULL;
for (j = 0; j nb_samples; j++) {
    v = (int)(sin(OutputStreamAudio.t) * 10000);
    for (i = 0; i < OutputStreamAudio.pAVCodecContext->channels; i++) *q++ = v;
    OutputStreamAudio.t     += OutputStreamAudio.tincr;
    OutputStreamAudio.tincr += OutputStreamAudio.tincr2;
}

frame->pts = OutputStreamAudio.next_pts;
OutputStreamAudio.next_pts  += frame->nb_samples;

return frame;


  


• doc/APIchanges : fill date & hash for AV_PIX_FMT_FLAG_BAYER

  20 mars 2017, par Clément Bœsch

  doc/APIchanges : fill date & hash for AV_PIX_FMT_FLAG_BAYER
  

  • [DH] doc/APIchanges