Recherche avancée

Sur d’autres sites (8873)

• ffmpeg continue streaming to twitch use case

  11 juin 2023, par Hassan M. Amin

  So I have a weird use case that I'm struggling with. Here's the situation :

  


  I have 2 video files, each 1 hour long. I want to stream the first file to twitch in a process, and when that file ends, the ffmpeg process will end and then I want to pickup the other video and continue the stream with that one.

  


  Why should the process end ? because after the first video ends, I need to check via code if the 2nd video is available to be streamed, otherwise I'll re-stream the first video again.

  


  I understand that this can be done via concatenating the two videos, but again, the 2nd video might not be available just yet so we need to end the process right after the 1st video ends streaming and check if it's there.

  


  This works great on YouTube, but the issue I'm facing is that Twitch specifically, when we start streaming the 2nd video, thinks that it's a totally new live stream and resets the live duration counter.

  


  Here's the million dollar question : How can we make twitch think that the data being sent from the 2nd video is the remainder of the first video ? I think it has something to do with the timestamps being sent ? any pointers or other way to look at this ?

  


  Tried updating the pts to be the previous video length + current pts

  


• avformat/matroskadec : Move WEBVTT code to mkv_parse_subtitle_codec()

  3 septembre 2023, par Andreas Rheinhardt

  avformat/matroskadec : Move WEBVTT code to mkv_parse_subtitle_codec()
  and also perform the remainder of the subtitle parsing directly
  
  after mkv_parse_subtitle_codec().
  Signed-off-by : Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
  

  • [DH] libavformat/matroskadec.c

• ffmpeg seek by position, not timestep [closed]

  23 octobre 2023, par user112346

  I'm implementing an application which requires extracting, say, 100 frames from a video stream, processing them, then extracting the next 100 and repeating this process until the end of the video.

  


  I've been doing this using :

  


  ffmpeg -i input.mkv -vf "select=gt(n\,$MYMIN_MINUS_ONE)*lt(n\,$MYMAX_PLUS_ONE)" -vsync 0 frames/frame%08d.png


  


  However, this approach encodes everything unnecessarily and is slow. I'm not sure if ffmpeg can seek based on the sequential frame position. Therefore, I changed the code to use timesteps by first doing :

  


  ffmpeg -itsscale $FRAME_RATE -i input.mkv -vf "fps=1" -q:v 1 -vsync vfr -c:v libx265 input_stretched.mkv


  


  Which stretches the input such that each frame in the original sequence corresponds to a different second. Then frame access is done by using :

  


  ffmpeg -ss $MYMIN -i input_stretched.mkv -frames:v $NFRAMES frames/frame%08d.png


  


  This solution seems to work and, after the initial encoding, access is much faster than using the filtergraph. I can probably alternatively divide the timestamps when accessing, instead of stretching the video, although I have not yet tested this.

  


  I'd like to know if there is a less "hacky" way of doing this, such as to use the frame sequence numbers directly, instead of converting to timesteps.