Recherche avancée

Sur d’autres sites (11419)

• Live streaming HLS via ffmpeg, How to force client to start playing from the beginning ? From 1st segment

  27 janvier 2021, par vedeojunky

  Is there a way, maybe via an ffmpeg option or flag, to force the client player to always start the playlist from the beginning when live streaming rather than the real time mid-stream ?

  



  Say the user comes in 1mn after the stream has started, rather than starting to watch at 1mn the player would start at the beginning of the video so minute zero.

  



  Here is my ffmpeg command :

  



  ffmpeg -f "screen capture" -s 1280x720 -r 30 -i :0.0+nomouse -f alsa -ac 2 -i pulse -async 30 -vcodec libx264 -pix_fmt yuv420p -acodec libfdk_aac -ar 44100 -b:a 64k -threads 0 -s 640x360 -f hls -g 1 -hls_time 1 -hls_list_size 1 -hls_allow_cache 0 /hls/#{@stream_name}/index.m3u8


  



  Thanks !

  


• Python-FFMPEG Corruption Problems

  11 juillet 2023, par Gabriel Ruben Guzman

  I'm repurposing some python code to generate gifs/mp4s showcasing nba player movements dot form. (With the 'frames' used in the gifs being generated by matplotlib).

  


  The repo comes with two different functions for generating the gifs, watch_play and animate_play. Both of which use python command line functionalities to run ffmpeg and generate the mp4s.
I've been able to use the watch_play succesfully, bot every time I try using animate_play, which according to the documention is meant to be significantly faster than watch play, I run into the error showcased here.(I printed the cmd string being passed into the pipe, in the hopes it would make debugging easier)

  


  I've tried generating gifs/mp4s of various size and added a decent bit of code to lessen the volume of data being processed. (I'm essentially repurposing the code just to generate clips, so I've been able to remove a lot of the pbp/tracking data logs to speed up the run time) But no matter what I've done, gotten some variation of the screenshotted error.

  


  pipe: : corrupt input packet in stream 0&#xA;[rawvideo @ 0x55ccc0e2bb80] Invalid buffer size, packet size 691200 < expected frame_size 921600&#xA;Error while decoding stream #0:0 : Invalid argument

  &#xA;

  The code for animate_play

  &#xA;

  def animate_play(self, game_time=None, length=None, highlight_player=None,
                 commentary=True, show_spacing=None):
    """
    Method for animating plays in game.
    Outputs video file of play in {cwd}/temp.
    Individual frames are streamed directly to ffmpeg without writing them
    to the disk, which is a great speed improvement over watch_play

    Args:
        game_time (int): time in game to start video
            (seconds into the game).
            Currently game_time can also be an tuple of length two
            with (starting_frame, ending_frame)if you want to
            watch a play using frames instead of game time.
        length (int): length of play to watch (seconds)
        highlight_player (str): If not None, video will highlight
            the circle of the inputed player for easy tracking.
        commentary (bool): Whether to include play-by-play commentary in
            the animation
        show_spacing (str) in ['home', 'away']: show convex hull
            spacing of home or away team.
            If None, does not show spacing.

    Returns: an instance of self, and outputs video file of play
    """
    if type(game_time) == tuple:
        starting_frame = game_time[0]
        ending_frame = game_time[1]
    else:
        game_time= self.start +(self.quarter*720)
        end_time= self.end +(self.quarter*720)
        length = end_time-game_time
        # Get starting and ending frame from requested 
        # game_time and length
        print('hit')
        print(len(self.moments))
        print(game_time)
        print(end_time)
        print(length)
        print(game_time+length)
        
        print(self.moments.game_time.min())
        print(self.moments.game_time.max())

        sys.exit()
        starting_frame = self.moments[self.moments.game_time.round() ==
                                      game_time].index.values[0]
        ending_frame = self.moments[self.moments.game_time.round() ==
                                    game_time + length].index.values[0]

    # Make video of each frame
    filename = "./temp/{game_time}.mp4".format(game_time=game_time)
    if commentary:
        size = (960, 960)
    else:
        size = (480, 480)
    cmdstring = ('ffmpeg',
                 '-y', '-r', '20',  # fps
                 '-s', '%dx%d' % size,  # size of image string
                 '-pix_fmt', 'argb',  # Stream argb data from matplotlib
                 '-f', 'rawvideo','-i', '-',
                 '-vcodec', 'libx264', filename)
    #print(pipe)
    #print(cmdstring)
    
    

    # Stream plots to pipe
    pipe = Popen(cmdstring, stdin=PIPE)
    print(cmdstring)
    for frame in range(starting_frame, ending_frame):
        print(frame)
        self.plot_frame(frame, highlight_player=highlight_player,
                        commentary=commentary, show_spacing=show_spacing,
                        pipe=pipe)
    print(cmdstring)
    pipe.stdin.close()
    pipe.wait()
    return self


  &#xA;

  The code for watch play

  &#xA;

  def watch_play(self, game_time=None, length=None, highlight_player=None,
               commentary=True, show_spacing=None):

    """
    DEPRECIATED.  See animate_play() for similar (fastere) method

    Method for viewing plays in game.
    Outputs video file of play in {cwd}/temp

    Args:
        game_time (int): time in game to start video
            (seconds into the game).
            Currently game_time can also be an tuple of length
            two with (starting_frame, ending_frame) if you want
            to watch a play using frames instead of game time.
        length (int): length of play to watch (seconds)
        highlight_player (str): If not None, video will highlight
            the circle of the inputed player for easy tracking.
        commentary (bool): Whether to include play-by-play
            commentary underneath video
        show_spacing (str in ['home', 'away']): show convex hull
            of home or away team.
            if None, does not display any convex hull

    Returns: an instance of self, and outputs video file of play
    """
    print('hit this point ')
    warnings.warn(("watch_play is extremely slow. "
                   "Use animate_play for similar functionality, "
                   "but greater efficiency"))

    if type(game_time) == tuple:
        starting_frame = game_time[0]
        ending_frame = game_time[1]
    else:
        # Get starting and ending frame from requested game_time and length
        game_time= self.start +(self.quarter*720)
        end_time= self.end +(self.quarter*720)
        length = end_time-game_time


        starting_frame = self.moments[self.moments.game_time.round() ==
                                      game_time].index.values[0]
        ending_frame = self.moments[self.moments.game_time.round() ==
                                    game_time + length].index.values[0]
    #print(self.moments.head(2))
    #print(starting_frame)
    #print(ending_frame)
    print(len(self.moments))
    # Make video of each frame
    title = str(starting_frame)+'-'+str(ending_frame)
    for frame in range(starting_frame, ending_frame):
        print(frame)
        self.plot_frame(frame, highlight_player=highlight_player,
                        commentary=commentary, show_spacing=show_spacing)
    command = ('ffmpeg -framerate 20 -start_number {starting_frame} '
               '-i %d.png -c:v libx264 -r 30 -pix_fmt yuv420p -vf '
               '"scale=trunc(iw/2)*2:trunc(ih/2)*2" {title}'
               '.mp4').format(starting_frame=starting_frame,title=title)
    os.chdir('temp')
    os.system(command)
    os.chdir('..')

    # Delete images
    for file in os.listdir('./temp'):
        if os.path.splitext(file)[1] == '.png':
            os.remove('./temp/{file}'.format(file=file))

    return self'


  &#xA;

• For converting video to frames, should I do client or server side processing ?

  23 mars 2024, par Tomas Marson

  Here is the thing, I have a Nodejs API that serves one video at a time when client request it.

  &#xA;

  The client (made in react) receives the video, which has no more than 15 seconds, watch it and decide if he wants to approve or deny it.&#xA;If he approves the video, it must be displayed in a sort of frames carousel, with one frame per second, so there is no more than 15 images/frames.

  &#xA;

  So the question is, should I do the conversion video-to-frames on client once he approves the video or should I do it on server-side and then request each frame (or streaming all frames with one request if possible) ?

  &#xA;

  Now, I'm doing the conversion on server with ffmpeg, but it seems tricky to send all the frames when the client already have them inside the video.

  &#xA;