Recherche avancée

Sur d’autres sites (4508)

• Seek in fragmented MP4

  15 novembre 2020, par Stefan Falk

  For my web-client I want the user to be able to play a track right away, without having to download the entire file. For this I am using a fragmented MP4 with the AAC audio coded (Mime-Type : audio/mp4; codecs="mp4a.40.2").

  


  This is the command that is being used in order to convert an input file to a fMP4 :

  


  ffmpeg -i /tmp/input.any \
  -f mp4 \
  -movflags faststart+separate_moof+empty_moov+default_base_moof \
  -acodec aac -b:a 256000 \
  -frag_duration 500K \
   /tmp/output.mp4


  


  If I look at this file on MP4Box.js, I see that the file is fragmented like this :

  


  ftyp
moov
moof
mdat
moof
mdat
..
moof
mdat
mfra


  


  This looks alright so far but the problem I am facing now is that it's not apparent to me how to start loading data from a specific timestamp without introducing an additional overhead. What I mean by this is that I need the exact byte offset of the first [moof][mdat] for a specific timestamp without the entire file being available.

  


  Let's say I have a file that looks like this :

  


  ftyp
moov
moof # 00:00
mdat 
moof # 00:01
mdat
moof # 00:02
mdat
moof # 00:03
mdat
mfra


  


  This file however, is not available on my server directly, it is being loaded from another service, but the client wants to request packets starting at 00:02.

  


  Is there a way to do this efficiently without me having to load the entire file from the other service to my server ?

  


  My guess would be to load [ftyp][moov] (or store at least this part on my own server) but as far as I know, the metadata stored in those boxes won't help me to find a byte-offset to the first [moof][mdat]-pair.

  


  Is this even possible or am I following the wrong approach here ?

  


• Convert ffmpeg yuv420p AVFrame to CMSampleBufferRef (kcvPixelFormatType_420YpCbCr8BiPlanarFullRange)

  21 juillet 2014, par user3272750

  I have a foscam ip camera and have access to the rtsp stream. I used DFURTSPPlayer to view the stream on my iOS device which works fine. I use a webrtc provider that lets me inject frames as CMSampleBufferRef in addition to directly reading from any of the on board cameras. I wish to use this to broadcast the IP camera stream over a secure webrtc session.

  The main loop in the DFURTSPPLayer checks if a frame is available and then converts into UIimage and sets it to an imageview.

  -(void)displayNextFrame:(NSTimer *)timer
  
  {
  
      NSTimeInterval startTime = [NSDate timeIntervalSinceReferenceDate];
  
      if (![video stepFrame]) {
  
          [timer invalidate];
  
          [playButton setEnabled:YES];
  
          [video closeAudio];
  
          return;
  
      }
  
      imageView.image = video.currentImage;
  
      float frameTime = 1.0/([NSDate timeIntervalSinceReferenceDate]-startTime);
  
      if (lastFrameTime<0) {
  
          lastFrameTime = frameTime;
  
      } else {
  
          lastFrameTime = LERP(frameTime, lastFrameTime, 0.8);
  
      }
  
      [label setText:[NSString stringWithFormat:@"%.0f",lastFrameTime]];
  
  }

  I’m trying to do something similar, but instead of (or in addition to) setting the UIImage I wish to also inject the frames into my webrtc service. This is an example where they use an avcapturesession. I believe I could do something similar to the runloop here and inject the frame (provided I can convert the yuv420p AVFrame into a CMSampleBufferRef :

  - (void) captureOutput:(AVCaptureOutput*) captureOutput
  
   didOutputSampleBuffer:(CMSampleBufferRef) sampleBuffer
  
          fromConnection:(AVCaptureConnection*) connection
  
  {
  
      self.videoFrame.frameBuffer = sampleBuffer;
  
  
  
      // IMPORTANT: injectFrame expects a 420YpCbCr8BiPlanarFullRange and frame
  
      //            gets timestamped inside the service.
  
      NSLog(@"videoframe buffer %@",self.videoFrame.frameBuffer);
  
      [self.service injectFrame:self.videoFrame];
  
  }

  Hence my question. Most of the questiosn on stack overflow involve going in the other direction (typically broadcasting on board camera input via rtsp). I’m a n00b as far as avfoundation/corevideo is concerned. I’m prepared to put in the groundwork if someone can suggest a path. Thanks in advance !

  Edit : After reading some more on this, it seems that most important step is a conversion from 420p to 420f.

• Do ffmpeg and ffprobe share the same code for metadata extraction ?

  3 décembre 2013, par user931392

  I want to make a simple transcoding service and I want to use ffprobe (because of the -print_format option) to get the metdata of a file (duration, width x height (if required), bit rate, etc) and ffmpeg for the transcoding itself.

  My question is :
  Is it safe to use the metadata from the ffprobe output to generate a command line for ffmpeg ?