Sur d’autres sites (8863)

• Merge multiple videos without using command line-FFMPEG [on hold]

  19 avril 2018, par Valgrind1691

  I am developing an android application which has video processing features involved.
  For video processing I have been working with FFMPEG library. As for now ffmpeg is directly being consumed by android and the processing is being done by
  framing commands and executing it using ffmpeg executable running in the background.
  This approach has two drawbacks :

  1) Video Processing time is huge even after trying ultrafast and other options to reduce processing

  2) APK size is large

  For solving two major problems we seek c/c++ native implementation for video processing by using the ffmpeg code.
  The C/C++ code will be cross compiled for android and iOS platforms and provided to mobile platform as a library (.so file in case of Android)
  The .so file will have

  1. jni for communication between Android(java) and c/c+

  2. c++ code for functionality implementation of video processing. All
     the video processing i.e scaling, transcoding, merging,
     cropping etc will be done by C/C++ code.

  I have a query around the C/C++ side , how can we use ffmpeg not to process using command line rather we can define algorithms for the process. Transcoding, muxing etc I have achieved looking into the examples of ffmpeg but I am still unable to figure out algorithm for merging of multiple(upto 6) videos in a single video file.
  I have looked into every question posted in this context and none helped.
  Can someone help me merge videos using source code of ffmpeg without using command line ? Any help is appreciated.

• Sampling video every N frames starting at n0 initial frame

  4 avril 2017, par Atcold

  Question

  I am trying to figure out how to sample a video stream x[n] every N frames, starting at frame n_i, i < N, so that I end up with N new videos of length len(x) / N.

  In formula this is simply : y_i[n] = x[n_i + n * N].

  Here there is a diagram of what I am trying to achieve :
  

  The greedy solution would be simply dumping the frames to a folder and then create new videos out of appropriately indexed frames.
  I was hoping there were some more elegant solution with ffmpeg since I have to process hundreds of video.

  Implementation

  Finally, I managed to write the final implementation, which I am reporting here for completeness.
  It does scale the minimum dimension to 256, does not process more than max_frames ; performs the sampling every k frames, send the first k - 1 samples to one folder and the k-th one to another one. It also set the output frame rate to the input average frame rate, since otherwise some videos will go at 120 Hz...

  k=5
  
  kk=$(awk "BEGIN{print 1/$k}")
  
  ffmpeg \
  
      -i $src_video_path \
  
      -an \
  
      -loglevel error \
  
      -filter_complex \
  
          "setpts=$kk*PTS, \
  
          scale=w=2*trunc(128*max(1\, iw/ih)):h=2*trunc(128*max(1\, ih/iw))[m]; \
  
          [m]select=n=$k:e=(mod(n\,$k)+1)*lt(n\,$max_frames) \
  
          $(for ((i=1; i<=$k; i++)); do
  
              echo -n "[a$i]"
  
          done)" \
  
      $(for ((i=1; i<$k; i++)); do
  
          echo -n "-r $fps -map [a$i] $dst_video_path/$i.mp4 "
  
      done
  
      echo -n "-r $fps -map [a$k] $val_video_path/$k.mp4"
  
      )

• lavc/bswapdsp : purge RISC-V V bswap32

  16 juillet 2023, par Rémi Denis-Courmont

  lavc/bswapdsp : purge RISC-V V bswap32
  This cannot beat the Zbb implementation, and it is unlikely that a real
  
  meaningful CPU design would support V and not Zbb. The best loop rewrite
  
  that I could come up with (4 shifts, 2 ands, 3 ors) is still 40% slower
  
  than Zbb.
  A proper faster vector implementation should be feasible with the
  
  cryptographic vector extensions, but that is a story for another time.
  

  • [DH] libavcodec/riscv/bswapdsp_init.c
  • [DH] libavcodec/riscv/bswapdsp_rvv.S