Advanced search

On other websites (5880)

• FFMPEG and DXVA2

  11 September 2017, by Bobnono

  I’ve made a player for IP camera (fullHD H264) with FFMPEG
  I code under windows with Qt.I need the lowest latency as possible and the best result I can get is done whith :

  • I got the packet with live555,
  • decode with ffmpeg
  • display it with SDL.

  But with "slow" CPU I have lot of latency if the bitrate is too high or it can’t handle 1080p.

  I really want to try hardware acceleration with DXVA2 to see if I can get better performance especially in slowest CPU.

  But i really don’t know how to proceed. I have to build FFMPEG with —enable-dxva2 and —enable-hwaccel=h264_dxva2 ? or Zeranoe build
  I have never use Direct X, do I need Direct X knowledge, or ffmpeg with —enable-dxva2 and —enable-hwaccel=h264_dxva2 care of everything ?

  Thank you

  Edit :

  Hello, I use this post to help me in my HW acceleration with FFMPEG
  Failed to execute: 0x80070057, when decoding video via ffmpeg with dxva2

  I decode the packet with

  avcodec_decode_video2(pCodecCtx, pFrame, &gotPic, &packet);

  and I’m getting my decoded frame with

  dxva2_retrieve_data_call(pCodecCtx, pFrame, &gotPic, &packet);

  I tried to convert it

  sws_scale(img_convert_affic, pFrame->data, pFrame->linesize, 0, pCodecCtx->height, pFrameRGB->data, pFrameRGB->linesize);

  and copy the buffer to a Qimage

  LastFrame=QImage(w,h,QImage::Format_RGB888);
  
  for(int y=0;ydata[0]+y*pFrameRGB->linesize[0],w*3);

  I got my picture but with lot of artefact and after few seconds I got a segmentation fault in dxva2_retrieve_data_call(pCodecCtx, pFrame, &gotPic, &packet);

  Edit :
  Problem solved i didn’t check if I got a picture before retreiving it.
  But the performance are very poor on my desktop cmputer (intel core i5 650, 4GB RAM, ATI HD5800) it consumes 5% more of CPU, 3 times more RAM, and lot of corrupt frame.
  It’s better on a newer computer, but soft decoder is better.
  dxva2_retrieve_data_call and sws_swale consume lot of time when hwaccel is enabled ...

  Is there a way to display NV12 picture in Qt (no sws_scale conversion) and there is a way to optimise dxva_retrieve_data_call ?

• FFMPEG and DXVA2

  5 February 2016, by Bobnono

  I’ve made a player for IP camera (fullHD H264) with FFMPEG
  I code under windows with Qt.I need the lowest latency as possible and the best result I can get is done whith :

  • I got the packet with live555,
  • decode with ffmpeg
  • display it with SDL.

  But with "slow" CPU I have lot of latency if the bitrate is too high or it can’t handle 1080p.

  I really want to try hardware acceleration with DXVA2 to see if I can get better performance especially in slowest CPU.

  But i really don’t know how to proceed. I have to build FFMPEG with —enable-dxva2 and —enable-hwaccel=h264_dxva2 ? or Zeranoe build
  I have never use Direct X, do I need Direct X knowledge, or ffmpeg with —enable-dxva2 and —enable-hwaccel=h264_dxva2 care of everything ?

  Thank you

  Edit :

  Hello, I use this post to help me in my HW acceleration with FFMPEG
  Failed to execute: 0x80070057, when decoding video via ffmpeg with dxva2

  I decode the packet with

  avcodec_decode_video2(pCodecCtx, pFrame, &gotPic, &packet);

  and I’m getting my decoded frame with

  dxva2_retrieve_data_call(pCodecCtx, pFrame, &gotPic, &packet);

  I tried to convert it

  sws_scale(img_convert_affic, pFrame->data, pFrame->linesize, 0, pCodecCtx->height, pFrameRGB->data, pFrameRGB->linesize);

  and copy the buffer to a Qimage

  LastFrame=QImage(w,h,QImage::Format_RGB888);
  
  for(int y=0;ydata[0]+y*pFrameRGB->linesize[0],w*3);

  I got my picture but with lot of artefact and after few seconds I got a segmentation fault in dxva2_retrieve_data_call(pCodecCtx, pFrame, &gotPic, &packet);

  Edit :
  Problem solved i didn’t check if I got a picture before retreiving it.
  But the performance are very poor on my desktop cmputer (intel core i5 650, 4GB RAM, ATI HD5800) it consumes 5% more of CPU, 3 times more RAM, and lot of corrupt frame.
  It’s better on a newer computer, but soft decoder is better.
  dxva2_retrieve_data_call and sws_swale consume lot of time when hwaccel is enabled ...

  Is there a way to display NV12 picture in Qt (no sws_scale conversion) and there is a way to optimise dxva_retrieve_data_call ?

• skip frame rendering if falling behind

  9 November 2016, by Brannon

  I have the code below that uses ffmpeg libraries (v3.1.4 with the Autogen wrapper) to render RTSP video in my application. The code works very well generally. However, the receptical.Write method is not particularly performant. On slow machines my video rendering starts to fall behind. Eventually my buffers fill up and I start to see video corruption. How can I change the code below to skip frames when it starts to fall behind? If there are multiple frames ready, I really only care to show the frame most recently available — this is live video after all. I believe that the avcodec_send_packet and avcodec_receive_frame methods are approximately 1-to-1.

  while (!token.IsCancellationRequested)
  
  {
  
      if (ffmpeg.av_read_frame(pFormatContext, pPacket) != 0)
  
      {
  
          // end of the stream
  
          ffmpeg.av_packet_unref(pPacket);
  
          ffmpeg.av_frame_unref(pDecodedFrame);
  
          break;
  
      }
  
  
  
      if (pPacket->stream_index != pStream->index || (pPacket->flags & ffmpeg.AV_PKT_FLAG_CORRUPT) > 0)
  
      {
  
          // this should never happen; we only subscribe to one stream
  
          // and I believe corrupt packets are automatically discarded
  
          ffmpeg.av_packet_unref(pPacket);
  
          ffmpeg.av_frame_unref(pDecodedFrame);
  
          continue;
  
      }
  
  
  
      var sendResult = ffmpeg.avcodec_send_packet(pCodecContext, pPacket);
  
      if (sendResult < 0)
  
      {
  
          // one of the possible results is a "buffer full", but I don't think that should happen as long as we call 1-to-1 receive_frame
  
          ffmpeg.av_packet_unref(pPacket);
  
          ffmpeg.av_frame_unref(pDecodedFrame);
  
          _logger.Warn("Failure in FFmpeg avcodec_send_packet: " + sendResult);
  
          break;
  
      }
  
  
  
      while (ffmpeg.avcodec_receive_frame(pCodecContext, pDecodedFrame) == 0)
  
      {
  
          var src = &pDecodedFrame->data0;
  
          var dst = &pConvertedFrame->data0;
  
          var srcStride = pDecodedFrame->linesize;
  
          var dstStride = pConvertedFrame->linesize;
  
          ffmpeg.sws_scale(pConvertContext, src, srcStride, 0, height, dst, dstStride);
  
  
  
          sbyte* convertedFrameAddress = pConvertedFrame->data0;
  
  
  
          int linesize = dstStride[0];
  
  
  
          if (receptical == null)
  
          {
  
              receptical = writableBitampCreationCallback.Invoke(new DetectedImageDimensions {Width = width, Height = height, Format = DetectedPixelFormat.Bgr24, Linesize = linesize});
  
          }
  
  
  
          var imageBufferPtr = new IntPtr(convertedFrameAddress);
  
          receptical.Write(width, height, imageBufferPtr, linesize);
  
  
  
          ffmpeg.av_frame_unref(pDecodedFrame);
  
      }
  
      ffmpeg.av_packet_unref(pPacket);
  
  }