Recherche avancée

Sur d’autres sites (9886)

• Non-Monotonous DTS error in current ffmpeg builds causing frame drops, but not on ffmpeg 1.2

  11 avril 2021, par pentinex

  I'm sorry to ask about such a frequently mentioned subject, but I looked through so many threads and I could never figure out what exactly is wrong, so I'm at the end of my wits. Basically, I want to concat a bunch of .ts files using ffmpeg. I use this command line :

  


  ffmpeg -allowed_extensions ALL -fflags +igndts -i chunklist.m3u8 -c copy "output.mkv"

  


  The .m3u8 file contains all the .ts parts. When I do this and it throws Non-Monotonous DTS errors, the result ends up having dropped frames where the DTS errors occur. I have tried multiple approaches :

  


  

  1. As you see above, I added "-fflags +igndts" at one point, though it doesn't help.

  


  2. I experimented with "-fflags +genpts" just out of desperation. Predictably, this made it worse.

  


  3. I tried muxing video and audio separately, to no avail.

  


  


  But weirdly enough, there is one "solution" : when I use ffmpeg 1.2, the resulting video plays smoothly. It doesn't come without issues. The resulting video always reports a frame rate of half of what it should be (14.98 when it should be 29.97). But it plays just like it should. Ideally though, I wouldn't want to use such an old version of ffmpeg, since I'm sure there are things in the background that may go wrong and I'm not even noticing those, so I'd hope I could get these kinds of results on current builds of ffmpeg.

  


  Does anybody have an idea ? I'm kind of losing my mind with this one.

  


• Streaming raw h264 video from Raspberry PI to server for capture and viewing [closed]

  24 juin 2024, par tbullers

  This is really an optimization question - I have been able to stream h264 from a raspberry pi 5 to a linux system and capture the streams and save them to .mp4 files.

  


  But I intend to run the video capture and sending on a battery powered Pi Zero 2 W and want to use the least amount of power to maximize battery life and still providing good video quality.

  


  I've explored many different configuration settings but am getting lost in all the options.

  


  This is what I run on the pi :

  


  rpicam-vid -t 30s --framerate 30 --hdr --inline --listen -o tcp://0.0.0.0:5000


  


  I retrieve this video from the more powerful Ubuntu server with :

  


  ffmpeg -r 30 -i tcp://ralph:5000 -vcodec copy video_out103.mp4


  


  It generally works but I receive lots of errors on the server side like this :

  


  [mp4 @ 0x5f9aab5d0800] pts has no valuee= 975.4kbits/s speed=1.19x
Last message repeated 15 times
[mp4 @ 0x5f9aab5d0800] pts has no valuee=1035.3kbits/s speed=1.19x
Last message repeated 15 times
[mp4 @ 0x5f9aab5d0800] pts has no valuee=1014.8kbits/s speed=1.18x
Last message repeated 9 times
[mp4 @ 0x5f9aab5d0800] pts has no valuee=1001.1kbits/s speed=1.17x
Last message repeated 7 times
[mp4 @ 0x5f9aab5d0800] pts has no value
Last message repeated 1 times
[out#0/mp4 @ 0x5f9aab5ad5c0] video:3546kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead : 0.120360%
size= 3550kB time=00:00:27.50 bitrate=1057.5kbits/s speed=1.18x

  


  Any suggestions on how to correct these errors ?

  


  Also any suggestions on how to make the video capture side more efficient ? Should I use a different codec ? (yuv instead of h264 ?) Would using UDP decrease overhead ? Can I improve video quality with the mode or hdr options ? What does denoise do ?

  


  With all the options available with these tools I think it's unlikely that I have a well thought out approach to capture and streaming. I'm hoping that people who are more familiar with this space might be able to provide some suggestions.

  


  Thank you !

  


  -tom

  


• Trouble trimming video by frame number with ffmpeg

  8 décembre 2023, par jgore200377

  I want to trim/split a video with ffmpeg using frames. The reason why is because I am using a Shot Transition Detection model which returns a probability of transition for every frame in the video.

  


  Using timestamps to cut the video has yielded bad results as the precision is not 100%.

  


  Ive tried this command which just outputs the entire video which is DEFINITELY not what I want

  


  ffmpeg -i video.mp4 -vf "trim=start_frame=100:end_frame=200,setpts=PTS-STARTPTS" -c:a copy output.mp4


  


  Ive also tried using python bindings with ffmpeg-python

  


  import ffmpeg

input_file = ffmpeg.input('video.mp4')
output_file = ffmpeg.output(input_file.trim(start_frame=1300, end_frame=1500), 'test_output.mp4')
ffmpeg.run(output_file)


  


  This doesnt work either and outputs a video with half of it being still with unpredictable length

  


  ---

  


  Ive visited some other sites but none seem to have this nailed down and it would be much appreciated if someone can answer how to use frames to trim/split a video with ffmpeg.