Recherche avancée

Sur d’autres sites (5108)

• avcodec/vc1 : Arm 64-bit NEON deblocking filter fast paths

  31 mars 2022, par Ben Avison

  avcodec/vc1 : Arm 64-bit NEON deblocking filter fast paths
  checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. Note that the C
  
  version can still outperform the NEON version in specific cases. The balance
  
  between different code paths is stream-dependent, but in practice the best
  
  case happens about 5% of the time, the worst case happens about 40% of the
  
  time, and the complexity of the remaining cases fall somewhere in between.
  
  Therefore, taking the average of the best and worst case timings is
  
  probably a conservative estimate of the degree by which the NEON code
  
  improves performance.
  vc1dsp.vc1_h_loop_filter4_bestcase_c : 10.7
  
  vc1dsp.vc1_h_loop_filter4_bestcase_neon : 43.5
  
  vc1dsp.vc1_h_loop_filter4_worstcase_c : 184.5
  
  vc1dsp.vc1_h_loop_filter4_worstcase_neon : 73.7
  
  vc1dsp.vc1_h_loop_filter8_bestcase_c : 31.2
  
  vc1dsp.vc1_h_loop_filter8_bestcase_neon : 62.2
  
  vc1dsp.vc1_h_loop_filter8_worstcase_c : 358.2
  
  vc1dsp.vc1_h_loop_filter8_worstcase_neon : 88.2
  
  vc1dsp.vc1_h_loop_filter16_bestcase_c : 51.0
  
  vc1dsp.vc1_h_loop_filter16_bestcase_neon : 107.7
  
  vc1dsp.vc1_h_loop_filter16_worstcase_c : 722.7
  
  vc1dsp.vc1_h_loop_filter16_worstcase_neon : 140.5
  
  vc1dsp.vc1_v_loop_filter4_bestcase_c : 9.7
  
  vc1dsp.vc1_v_loop_filter4_bestcase_neon : 43.0
  
  vc1dsp.vc1_v_loop_filter4_worstcase_c : 178.7
  
  vc1dsp.vc1_v_loop_filter4_worstcase_neon : 69.0
  
  vc1dsp.vc1_v_loop_filter8_bestcase_c : 30.2
  
  vc1dsp.vc1_v_loop_filter8_bestcase_neon : 50.7
  
  vc1dsp.vc1_v_loop_filter8_worstcase_c : 353.0
  
  vc1dsp.vc1_v_loop_filter8_worstcase_neon : 69.2
  
  vc1dsp.vc1_v_loop_filter16_bestcase_c : 60.0
  
  vc1dsp.vc1_v_loop_filter16_bestcase_neon : 90.0
  
  vc1dsp.vc1_v_loop_filter16_worstcase_c : 714.2
  
  vc1dsp.vc1_v_loop_filter16_worstcase_neon : 97.2
  Signed-off-by : Ben Avison <bavison@riscosopen.org>
  
  Signed-off-by : Martin Storsjö <martin@martin.st>
  

  • [DH] libavcodec/aarch64/Makefile
  • [DH] libavcodec/aarch64/vc1dsp_init_aarch64.c
  • [DH] libavcodec/aarch64/vc1dsp_neon.S

• How do I convert a 3D SBS dual-fisheye image to 3D SBS dual-equirectangular with only open-source tools ? [closed]

  21 octobre 2024, par Ethan T

  Canon has released a new 3D VR lens for their RF mount and I recently rented it to see what it's like to work with. Unfortunately, they charge a subscription fee for their conversion software, which I've also found to be inconveniently limited in real-world use cases. Thus I'm interested in an open-source approach to converting the video captured by this lens using something like ffmpeg.

  


  The video is dual-fisheye and the Canon tool produces dual-equirectangular side-by-side video. I would like to perform this same conversion in ffmpeg or a similarly-powerful open-source tool.

  


  Related questions exist but aren't quite correct. Most dual-fisheye input seems to be used to create 360-degree 2D video, not 3D SBS.

  


• How do I render a video from a list of time-stamped images ?

  31 mai 2022, par piedar

  I have a directory full of images following the pattern <timestamp>.png</timestamp>, where <timestamp></timestamp> represents milliseconds elapsed since the first image. input.txt contains a list of the interesting images :

  



  file '0.png'
file '97.png'
file '178.png'
file '242.png'
file '296.png'
file '363.png'
...


  



  I am using ffmpeg to concatenate these images into a video :

  



  ffmpeg -r 15 -f concat -i input.txt output.webm


  



  How do I tell ffmpeg to place each frame at its actual position in time instead of using a constant framerate ?