Recherche avancée

Sur d’autres sites (13496)

• ffmpeg : is there a fast way for extracting several thumbnails from a video without parsing the video from the beginning every time ?

  18 mars, par archie

  I tried several ways for extracting sample frames from a video file with ffmpeg. I found out that the fastest way is by placing the following command in a loop :

  


  ffmpeg -ss $frame_time -i "$input_video" -frames:v 1 -vf scale=256:-1 "$Work_dir/thumb$thumb_index.jpg"


  


  (I have omitted the parts of the command that are not relevant to the question, such as drawtext, hide_banner, loglevel). The variables frame_time and thumb_index are initialized before the loop and incremented by a fixed amount at every step : +1 for thumbs_index and $duration/25 for frame_time. I read in the ffmpeg documentation (https://trac.ffmpeg.org/wiki/Seeking) that having the -ss part before the -i part is very fast because the input is parsed by keyframe. For the same reason, the loop containing the command above is also much faster than commands based on "-vf fps=$thumbs_number/$duration".

  


  In fact, the code works pretty well. However, I can't escape a feeling of programming discomfort, because ffmpeg is called several times for each video file, and every time it has to parse the file from the beginning. I mean, if I had a function doing the same as the command above — it would parse a video from the beginning to search for a frame at a certain time — calling it n times to extract a regular sequence of n frames would be bad programming. I should change the function to parse the video file once and get all the frames I need in a single pass.

  


  My question is : is there a fast way for having ffmpeg parse a video file a single time, searching by keyframes and extracting a given number of frames at a given distance from one another ? I am ready to take No for an answer, but one who is deeper into ffmpeg than I am might know a way. Thanks

  


• Seeking issues on larger video files on HTML page (ffmpeg)

  5 juillet 2023, par Xenons

  After testing out numerous of "html video players" and even making our own making sure to wait for events such as player.oncanplaythrough to fire before even attempting to start videos
we've had the same issue for months now.

  


  Mainly we offer 3 resolutions for same video (720, 1080, 4k)
and out of the 3, 2 have the seeking issue where trying to skip to any time or even trying to fast forward by x amount of seconds with arrow keys would just bring the video back to 00:00 time
this only happens on 1080 and 4k resolutions and after refreshing the page it just works as normal afterwards

  


  we've tried pretty much anything we could think of to combat this but with no luck

  


  ffmpeg line for 1080 and 4k (these files are webm but same happened for mp4 on these resolutions with same encoding)

  


  -c:v libvpx-vp9 -pix_fmt yuv420p -crf 24 -b:v 0 -c:a libopus -b:a 128k -movflags +faststart


  


  ffmpeg line for 720 (mp4 file)

  


  -c:v libx264 -preset slow -pix_fmt yuv420p -crf 20 -b:a 128k


  


  To be quite frank at this point I am at a loss and without many ideas left so I am turning to stack overflow in case someone might have an idea on what would cause this

  


  I am contemplating to attempt perhaps loading videos as blobs and playing them that way, not sure if this would fix it but it does bring the load times up drastically so I would save it as last resort

  


  Thus far I have tried two html video players (plyr and videojs) and after that I just wrote my own "player"
of course it's just regular html and js but that's what makes me even more confused as there is nothing that should hinder the files in any way from working as intended

  


  we've also tried changing ffmpeg encoding a few times but to no avail either.

  


  EDIT :
The issue was caused by cloudflare caching, we simply made two rules to exclude webm and mp4 url's from being cached and that completely fixed all of our seeking issues

  


• ffmpeg FPS information mismatch with the video

  8 décembre 2017, par Adorn

  I have a bunch of videos with some statistics of what is happening inside a video. One such piece of information is given in terms of time of the video in seconds up to one decimal point.

  To get the FPS of a video, I am using ffmpeg -i

  But when I manually compute one particular frame’s time using given FPS, it does not match.

  For example, ffmpeg outputs FPS = 30.

  I look at the video statistics, the frame at the 156.8 = 2.368 has to be
  4704'th frame. I open the video using ’skvideo’, read all the frames, and view the 4704'th frame. It is some frame around time 2.12 ?. I checked multiple such instances in multiple videos and this is a common behavior.

  I do not understand why this is so and how can I get around the problem ?

  As such I am not bounded by ffmpeg. Skvideo is being used to read the videos. I tried opencv, as of now it does not work with VideoCapture, and reinstalling it is costly for me time wise. But I guess ’opencv/skvideo’ should not matter, one can count the frames manually as well.

  So, in the solution, I am looking out for -

  1. Given timestamps of inside of a video, how can I find a frame of that particular time location ?

  2. In case someone might have already worked on this, this is related to THUMOS dataset. I am on Ubuntu 16.04

  EDIT_1

  Actually I can be more specific as it is a publicly available data. The time bounds are of an important activity. For example, in a video, when does basketball dunk occurs ? It is given in pairs - [start end]. Some videos have multiple activities, some have only one.

  Here is a sample video, and following are the activity times.

  [[  16.5, 20.8],
  
  [  26.6, 32.2],
  
  [  34.8, 42.1],
  
  [  47.8, 50.0],
  
  [  58.1, 62.9],
  
  [  65.6, 67.2],
  
  [  68.5, 74.0],
  
  [  76.4, 78.3],
  
  [  78.7, 79.8],
  
  [  80.8, 82.1],
  
  [  85.0, 87.3],
  
  [  90.1, 91.4],
  
  [  98.5, 100.3]]

  I also tried checking manually, 32.87 FPS "almost" works for few videos but not for all. and almost means it is off by 10 frames. This is a huge difference for my task, and I need exact frame.

  Also, there has to be some way, because it can be visually observed with multiple video players that times in the dataset are correct.