Recherche avancée

Sur d’autres sites (7920)

• FFMPEG bottleneck in relaying data from a dshow camera to stdout PIPE without any processing or conversion

  19 août 2020, par koonyook

  I have a USB camera (FSCAM_CU135) that can encode the video to MJPEG internally and it supports DirectShow. My goal is to retrieve the binary stream of the encoded video as it is (without decoding or preview) and send it to my program for further processing.

  


  I choose to use FFMPEG to read the MJPEG stream and pipe to stdout so that I can read it using Python's subprocess.Popen .

  


  ffmpeg -y -f dshow -vsync 2 -rtbufsize 1000M -video_size 1920x1440 -vcodec mjpeg -i video="FSCAM_CU135" -vcodec copy -f mjpeg pipe:1


  


  At this resolution, the camera is able to capture and transmit at 60 fps.
In this case, I expect FFMPEG to pass the data as fast as possible with no calculation.
With the output of FFMPEG I can tell how fast it moves the data from rtbuffer to the output pipe.

  


  With just one camera, FFMPEG works with no problem and move the data at 60 fps.
However, when I run 2 cameras simultaneously, the cameras still generate data at 60 fps but FFMPEG can only move the data around 55 fps. This means that I am unable to consume the video in realtime and the buffer consumption will be larger over time.

  


  I guess that FFMPEG didn't just simply move the data but did some processing such as searching for the beginning, the end, and the timestamp of each video frame so that it can count frames and report.
Is there a way to force FFMPEG to not doing those things and focus on passing the data only to make it faster ?

  


  If I purely use directshow API without FFMPEG, can it be faster ?

  


• Encoding a Data Stream Alongside Video in FFMPEG

  11 octobre 2016, par AdmiralJonB

  I’m wanting to encode some proprietary data (it’s a serialised unsigned 64-bit integer per frame) into a video container (mp4) as a data stream, but I have been unable to find any instructions/tutorials of anyone doing this.

  The only thing I’ve been able to find is this, which describes potentially how to create a data stream (the user had no success apparently
  https://lists.libav.org/pipermail/ffmpeg-user/2006-November/005070.html)

  This is my current code for creating a stream :

      ff_data_stream = avformat_new_stream(ff_output_context, NULL);
  
  
  
      ff_data_stream->codec->codec_type = AVMEDIA_TYPE_DATA;
  
      ff_data_stream->codec->codec_id = AV_CODEC_ID_NONE;
  
      ff_data_stream->codec->flags |= CODEC_FLAG_GLOBAL_HEADER;

  But then when I’m calling avformat_write_header it errors with the following output to console :

  [mp4 @ 0x7fff68000900] track 1 : could not find tag, codec not
  currently supported in container

  So my questions are as follows :

  • Is it possible to create a data stream with an mp4 container ? If not, are there any containers that do ?
    • This might not be the right way to do this, but I have not yet come across any method of doing so.
  • If so, how can I configure the stream correctly ? (Whether it’s for this container or another)
  • Would one then use an AVPacket when writing to file ? And write it into the file by using av_interleaved_write_frame ?

  Thanks

• Displaying YUV420 data using Opengles shader is too slow

  28 novembre 2012, par user1278982

  I have a child thread called A to decode video using ffmpeg on iPhone 3GS, another thread called B to display yuv data, in thread B, I used glSubTexImage2D to upload Y U V textures, and then convert yuv data to RGB in shader, but the frame rate in the decode thread is only 15fps.Why ?

  Update :
  The frame size is 720 * 576.
  I also found something interesting that if I didn't start the thread displaying the YUV data, the frame rate calculated in the decode thread is 22 fps,otherwise 15 fps.So I think that my displaying method must be inefficient.the code as below.

  I have a callback in the decode thread :

  typedef struct _DVDVideoPicture
  
  {
  
     char *plane[4];
  
     int iLineSize[4];
  
  }DVDVideoPicture;
  
  
  
  void YUVCallBack(void *pYUVData, void *pContext)
  
  {
  
     VideoView *view = (VideoView *)pContext;
  
     [view.glView copyYUVData:(DVDVideoPicture *)pData];
  
     [view calculateFrameRate];
  
  }

  The copyYUVData method extract the y u v planes separately. The following is displaying thread method.