Recherche avancée

Sur d’autres sites (8447)

• Ffmpeg burnt in subtitles out of sync when converting to hls

  19 mai 2020, par user1503606

  I have a file that has subtitles burn into it and they are perfectly in sync.

  



  Here is the file. https://983yqbz442.s3.amazonaws.com/little-mermaid-captions.mp4

  



  I run this command to convert to hls and it creates the .ts files and the .vtt files.

  



  ffmpeg -i little-mermaid-captions.mp4 -profile:v baseline -level 3.0 -s 640x360 -start_number 0 -hls_time 10 -hls_list_size 0 -f hls index.m3u8


  



  I also then create a master.m3u8 file in the same folder with the following.

  



  #EXTM3U
#EXT-X-VERSION:3
#EXT-X-INDEPENDENT-SEGMENTS
#EXT-X-STREAM-INF:BANDWIDTH=800000,RESOLUTION=640x360,SUBTITLES="subtitles"
index.m3u8
#EXT-X-MEDIA:TYPE=SUBTITLES,GROUP-ID="subtitles",NAME="English",DEFAULT=YES,AUTOSELECT=YES,FORCED=NO,LANGUAGE="ENG",URI="index_vtt.m3u8"


  



  Now if I play the master.m3u8 file the subtitles are now out of sync and are about 1 sec to quick. I understand this is probably a setting I am missing through FFmpeg but really stuck on this and would appreciate any insight.

  



  Thanks

  



  More info.

  



  Here is a link to the direct .m3u8 this can be opened in Safari.

  



  https://983yqbz442.s3.amazonaws.com/hlstests/master.m3u8

  



  The generated vtt file is here.

  



  https://983yqbz442.s3.amazonaws.com/hlstests/subs-0.vtt

  



  If you look at the start of the .vtt file you will see this.

  



  WEBVTT

00:06.840 --> 00:10.320
once long ago in the deep blue below


  



  It should start at 00:06.840 but when playing the .m3u8 file in Safari you should see it starts at around 5 seconds not 6 about a second too early.

  


• ffmpeg RTSP streams to RGB24 using GPU

  31 janvier 2020, par jerq

  I have successfully output RTSP stream to RGB24 pipe but I notice the CPU usage is still high (20-45%) despite using -hwaccel cuvid and -vcodec h264_cuvid (without hardware acceleration it can go up to 100-300% ; based on "top" command on Linux).

  The current code I am using is :

  ffmpeg
  
  .input(*rtsp_add*, rtsp_transport='tcp', fflags='nobuffer', flags='low_delay', hwaccel='cuvid', vcodec='h264_cuvid', vsync=0)
  
  .output('pipe:', format='rawvideo', pix_fmt='rgb24')
  
  .run_async(quiet=False, pipe_stdout=True)

  The moment i use -vcodec h264_nvenc, I get a bunch of error as I believe it is incompatible with RGB24. Using the command "ffmpeg -pix_fmts", I notice almost nothing falls under the "Hardware accelerated format". I do not mind working in YUV420p, YUV444p, RGB24 etc... Does this mean that I am unable to output images into the pipe if I choose to use hardware encoding ?

  I would prefer to write the output to a pipe as compared to writing it to a physical file and reading it again to manipulate the data for performance. The option to output to a in-memory JPG or PNG is also good.

• C/C++ examples for HW acceleration with FFMPEG on Android (JNI)

  12 février 2024, par Ramil Galin

  Are there any good C/C++ tutorials or code examples for how can one write hardware accelerated decoding on Android (JNI) ?

  


  I found related questions here on stackoverflow (here and here), but would like to know if there is something more contemporary and related to Android.