Recherche avancée

Sur d’autres sites (7483)

• avformat/mpegtsenc : do not include adaptation field in teletext TS packets

  21 octobre 2021, par Alex Shumsky

  avformat/mpegtsenc : do not include adaptation field in teletext TS packets
  From ETSI EN 300 472 V1.3.1 (2003-05) Specification for conveying ITU-R System
  
  B Teletext in DVB bitstreams :
  4.1 Transport Stream (TS) packet format
  
  The standard TS packet syntax and semantics are followed, noting the following
  
  constraint :
  
   adaptation_field_control only the values "01" and "10" are permitted.
  Some set top boxes (Motorola, Arris, Zyxel) refuse non-conforming packets.
  Signed-off-by : Alex Shumsky <alexthreed@gmail.com>
  
  Signed-off-by : Marton Balint <cus@passwd.hu>
  

  • [DH] libavformat/mpegtsenc.c

• OS X Mavericks broken OpenCV VideoCapture frame numbers ?

  20 novembre 2013, par MattP

  Fresh install of Mavericks, managed to install ffmpeg via homebrew and compile the latest OpenCV at time of writing. My code is written in Python (though this shouldn't affect this issue) and basically navigates a video file using the VideoCapture class with the properties set frame position and get current position.

  The issue is, on prior versions of OS X, various Linux and Windows boxes, this works without issue. Each frame number is of the format i.0, where i is the current frame number (why it is a float anyway is confusing, alas it's how OpenCV gives it back). However, on this setup I get frame progressions like :

  0.0 -> 1.3972597329753 -> 2.999999999 -> 3.9999999 -> 5.0 -> 6.252323552

  This makes using the properties for getting / setting the current frame position impossible and, as one would imagine, causes the automated playback thereof to fail. I can only assume ffmpeg is reporting things odd such that OpenCV is incorrectly calculating frames, but have been unable to rectify this.

  Has anyone any insight into the problem, and/or possible solution ? Thanks.

• Random access to video frame by its id

  22 août 2018, par Xmiler

  I have a 24 hour video and would like to organize storage of some meta information (particularly bounding boxes of objects) for arbitrary frame. It implies that I must have a reliable random access to frame by its id, practically even to the range of frames in a way (id_start, id_end).

  Once again, it is critically to get the same frame when I fill meta storage and when I will get meta later (in general in another environment - ffmpeg version, codecs etc). Otherwise I will put bounding box on the wrong frame.

  So there ways I found

  1. ffmpeg -i in.mp4 -vf select='between(n,x,y)' -vsync 0 frames%d.png
     The problem is it works too slow. Also I don’t know is it reliable in sense of reproducing.

  2. ffmpeg -ss pkt_pts_time -i in.mp4 -vframes n output_%03d.png
     where pkt_pts_time got from ffprobe -show_frames in.mp4 -select_streams v
     The problem is pkt_pts_time not clear at all. Does pkt_pts_time refer to -ss parameter ?

  Actually the question is - does internal meta information of frame (say h.264) contain frame id ? I didn’t find way to get it.

  Now I am thinkig to store png md5sum for each T’s frame (anchor). Later I will be able to check consistency using them.

  Can anyone suggest something better ?