Recherche avancée

Sur d’autres sites (10882)

• Why does ffmpeg have bigger latency on dark images ?

  19 novembre 2017, par doodoroma

  I have a c# application to stream real camera images using ffmpeg. The input images are in raw, 8-bit gray-scale format. I created an ffmpeg stream using the standard input to feed the images and send the output packages to websocket clients.

  I start an external ffmpeg process using this config :

  -f rawvideo -pixel_format gray -video_size " + camera.Width.ToString() + "x" + camera.Height.ToString() + " -framerate 25 -i - -f mpeg1video -b:v 512k -s 320x240 -

  Typical image size is 1040*1392 pixels

  I display the stream on the browser, using jsmpeg library

  This works with a reasonable latency ( 500ms on localhost), but when the camera image is really dark (black image), the latency is extremely big ( 2-3 seconds on localhost). When there is something bright on the image again after a black period, it takes 2-3 seconds to "synchronize".

  I was thinking that black images are really easy to compress and will generate really small packages, jsmpeg has almost no information to display and wait until a complete data package arrives, but I couldn’t prove my theory.

  I played with ffmpeg parameters like bitrate and fps but nothing has changed.

  Is there any settings which I could try ?

• FFmpeg programming : exr image too dark [on hold]

  2 juillet 2017, par Francis Tesla

  Using the fantastic FFmpeg library to load images all exr files appears dark.
  I am almost sure that’s involve converting the resulting AVFrame or AVPicture from linear to sRGB but I don’t know how to do it.

  Here is how the EXR file is loaded into an AVPicture :

  AVFrame *frame = av_frame_alloc();
  
  avcodec_decode_video2(codecCtx, frame, &frameFinished, &packet);
  
  
  
  ... 
  
  
  
  AVPicture       picture;
  
  avpicture_alloc(&picture, dest_format, dest_width, dest_height);
  
  
  
  SwsContext *swsCtx = NULL;
  
  
  
  swsCtx = sws_getContext(source_width, source_height, source_format, dest_width, dest_height, dest_format, SWS_POINT, NULL, NULL, NULL);
  
  
  
  if (swsCtx == NULL) {
  
      qDebug() << "error calling sws_getContext";
  
  }
  
  
  
  sws_scale(swsCtx, frame->data, frame->linesize, 0, frame->height, picture.data, picture.linesize);
  
  
  
  QImage image(dest_width, dest_height, QImage::Format_ARGB32);
  
  
  
  for (int y = 0; y < dest_height; ++y) {
  
      memcpy(image.scanLine(y), picture.data[0] + y * picture.linesize[0], picture.linesize[0]);
  
  }
  
  
  
  av_frame_free(&frame);
  
  sws_freeContext(swsCtx);

  Regards,

  Francis

• Revert "lavu/atomic : add support for the new memory model aware gcc built-ins"

  7 mars 2017, par James Almer

  Revert "lavu/atomic : add support for the new memory model aware gcc built-ins"
  This reverts commit faa9d2982969c999ab0e443a226eff116f7f8e4b.
  This change became superfluous when support for C11 atomics was introduced.
  
  Reverting it will make the removal of this implementation in an upcoming
  
  merge conflict free.
  Reviewed-by : wm4 <nfxjfg@googlemail.com>
  
  Signed-off-by : James Almer <jamrial@gmail.com>
  

  • [DH] configure
  • [DH] libavutil/atomic_gcc.h