Recherche avancée

Sur d’autres sites (6407)

• ffmpeg : Turning images/audio into video with ffmpeg

  7 mai 2018, par melanie93

  I am using ffmpeg to turn a sequence of images (img001.png, img002.png, and so on) into a video (output.mp4) using the following command :

  ffmpeg -r 1/5 -i img%03d.png  -r 25 -qscale:v 2 -shortest -codec:a copy output.mp4

  The result is a video that displays every input image for five (5) seconds.

  1. Is it possible to have ffmpeg parse the filename paths and timings from a file ? I tried the slideshow tutorial on the official ffmpeg website but the output displayed only the last image, briefly, at the end of the video.
  2. Is it possible to bundle audio files with those settings ?

  For example :

  file 'image001.png'
  
  file 'sound001.wav'
  
  duration 5
  
  file 'image002.png'
  
  file 'sound002.wav'
  
  duration 2
  
  file 'image003.png'
  
  file 'sound003.wav'
  
  duration 3

  Image001 is displayed for five(5) seconds while sound001 is being played back and so on.

• matroskadec : introduce resync function.

  28 mai 2013, par Sean McGovern

  matroskadec : introduce resync function.
  This allows handling matroska files with errors.
  
  Fixes test4.mkv and test7.mkv from the official Matroska test suite,
  
  and by extension Bugzilla #62.
  Based on a patch by Reimar Doffinger <Reimar.Doeffinger@gmx.de>
  Signed-off-by : Anton Khirnov <anton@khirnov.net>
  

  • [DBH] libavformat/matroskadec.c

• How to duplicate frames if appsrc is pushing too slow to keep frame rate up ?

  18 juillet 2023, par Jan

  I am using Java with GStreamer wrapped by OpenCV. The application does the following :

  


  


  1. One GStreamer pipeline, that receives frames from an IP Camera (RTSP stream with 30fps) and outputs OpenCV Mats
"rtspsrc location=rtsp://localhost:8554/looping-sample latency=500 protocols=udp buffer-mode=auto ! rtph264depay ! h264parse ! avdec_h264 max-threads=2 ! videoconvert ! video/x-raw,format=BGR ! appsink max-buffers=0 drop=true sync=false"

     


  


  2. On those Mats, deep learning models infer and draw stuff on it (This might be slow, like 100ms or sth)

     


  


  3. Another GStreamer pipeline wrapped by OpenCV at the 30fps (extracted by the RTSP stream), that receives those Mats and publishes an RTMP stream for a MediaServer
"appsrc ! videoconvert ! nvh264enc bitrate=3000 ! h264parse ! flvmux ! rtmpsink location=test"

     


  


  


  Now, the problem is, that the OpenCV VideoWriter is configured to be 30fps but the appsrc is only pushing at e.g. 7fps, so the RTMP stream will not be classified as valid by the MediaServer.
Previously, I used OpenCV and FFFMPEG and the default VideoWriter just duplicated the last frame to keep up the pre-defined framerate, so the stream is laggy but works.

  


  How can I achieve this with a GStreamer pipeline ?

  


  TLDR : The OpenCV VideoWriter is configured to be 30fps but appsrc is only pushing with 7-10fps, how can I adaptively duplicate frames to keep the 30fps in the RTMP stream sink ?

  


  I tried stuff like adding "appsrc ! videorate ..." to the pipeline, so that videorate keeps up the framerate, but that did not work