Recherche avancée

Sur d’autres sites (9580)

• Fastest way to extract raw Y' plane data from Y'Cb'Cr encoded video ?

  20 février 2024, par memeko

  I have a use-case where I'm extracting I-Frames from videos and turning them into perceptual hashes for later analysis.

  


  ⠀

  


  I'm currently using ffmpeg to do this with a command akin to :

  


  ffmpeg -skip_frame nokey -i 'in%~1.mkv' -vsync vfr -frame_pts true -vf 'keyframes/_Y/out%~1/%%06d.bmp'

  


  and then reading in the data from the resulting images.

  


  ⠀

  


  This is a bit wasteful as, to my understanding, ffmpeg does implicit YUV -> RGB colour-space conversion and I'm also needlessly saving intermediate data to disk.

  


  Most modern video codecs utilise chroma subsampling and have frames encoded in a Y'C_bC_r colour-space, where Y' is the luma component, and Cb Cr are the blue-difference, red-difference chroma components.

  


  Which in something like YUV420p used in h.264/h.265 video codecs is encoded as such :

  


  

  


  Where each Y' value is 8 bits long and corresponds to a pixel.

  


  ⠀

  


  As I use gray-scale data for generating the perceptual hashes anyway, I was wondering if there is a way to simply grab just the raw Y' values from any given I-Frame into an array and skip all of the unnecessary conversions and extra steps ?

  


  (as the luma component is essentially equivalent to the grayscale data i need for generating hashes)

  


  I came across the -vf 'extractplanes=y' filter in ffmpeg that seems like it might do just that, but according to source :

  


  


  "...what is extracted by 'extractplanes' is not raw data of the (for example) Y plane. Each extracted is converted to grayscale. That is, the converted video data has YUV (or RGB) which is different from the input."

  


  


  which makes it seem like it's touching chroma components and doing some conversion anyway, in testing applying this filter didn't affect the processing time of the I-Frame extraction either.

  


  ⠀

  


  My script is currently written in Python, but I am in the process of migrating it to C++, so I would prefer any solutions pertaining to the latter.

  


  ffmpeg seems like the ideal candidate for this task, but I really am looking for whatever solution that would ingest the data fastest, preferably saving directly to RAM, as I'll be processing a large number of video files and discarding I-Frame luma pixel data once a hash has been generated.

  


  I would also like to associate each I-Frame with its corresponding frame number in the video.

  


• Android Video color effect issue using FFMPEG [on hold]

  8 juin 2016, par umesh mishra

  I m facing one problem. when we use library for effect that i have mentioned below video is created only 1/3 of actual video size. please tell me what is issue. and also doesn’t work on marshmallow.
  https://github.com/krazykira/VidEffects/wiki/Permanent-video-effects

• Splitting audio by vocal / voice

  1er octobre 2020, par ML85

  I am working with audio file using webrtcvad and pydub. The split of any fragment is by silence of the sentence.
Is there any way by which the split can be done at each vocal (after each spoken word) ?
If librosa/ffmpeg/pydub has any feature like this, can split is possible at each vocal ? but after split, I need start and end time of the vocal exactly what that vocal part has positioned in the original file.
One simple solution or way to split by ffmpeg is also defined by :

  


  https://gist.github.com/vadimkantorov/00bf4fbe4323360722e3d2220cc2915e

  


  but this is also splitting by silence, and with each padding number or the frame size, the split is different. I am trying split by vocal.