Recherche avancée

Sur d’autres sites (8394)

• x86/vvc_alf : avoid overwriting for non-16 aligned widths

  20 juin 2024, par Nuo Mi

  x86/vvc_alf : avoid overwriting for non-16 aligned widths
  Previously, the code allowed overwriting on 16-aligned blocks, which was suitable when there were
  
  no picture's virtual boundaries because both CTU sizes and strides were 16-aligned. However, with
  
  picture's virtual boundaries, each CTU is divided into four ALF blocks, leading to potential issues
  
  with overwriting later CTUs.
  In cases involving picture virtual boundaries, each ALF block is 8-pixel aligned.
  
  For luma, we consistently ensure an 8-aligned width. For chroma in 4:2:0 format,
  
  we need to account for a 4-aligned width.
  

  • [DH] libavcodec/x86/vvc/vvc_alf.asm

• MPEG DASH SRD : How to properly re-assemble multiple tiles into a single tile [closed]

  21 juin 2024, par ATrashInTheWorld

  I am trying to build a web player that is able to stream MPEG DASH, but with SRD information in the manifest. I have a problem at the "gluing" of the tiles part.

  


  How the data is set :

  


  I have a video of a resolution of XxY, which I've divided in tiles of NxM, meaning that I now have the NxM videos of the same duration as the original one, but which is reduced to a smaller resolution, a portion from the original (seeing only the upper left corner for example).

  


  I have fragmented those tiles for the MPEG DASH streaming format, meaning an init.mp4 and the m4s fragments for each tile. To note that I have the tiles in multiple bitrate, in order to imitate some adaptive streaming.

  


  The "gluing" part :

  


  The gluing part consists of gluing the tile togheter into 1 big tile, aka the original video. The SRD information allows me to know where each tile should be position to reconstruct the original video shape. Once I have the spatial information from the manifest, I use this command to glue the tiles togheter :

  


  ffmpeg.exe -i v1 -i v2 -i v3 -i .v4 -i v5 -i v6 -i v7 -i v8 -i v9 -filter_complex "[0:v][1:v][2:v][3:v][4:v][5:v][6:v][7:v][8:v]xstack=inputs=9:layout=0_0|w0_0|w0+w1_0|0_h0|w3_h0|w3+w4_h0|0_h0+h3|w6_h0+h3|w6+w7_h0+h3" out.mp4


  


  The command above allows me to "glue" together tiles from a grid of 3x3 tiles, into 1 big tile.

  


  The web streaming method :

  


  To stream, I use JavaScript MediaSource, and I am appending the fragments to the buffer continuously, like this reference

  


  The issue :

  


  The issue is that the FFMPEG gluing command only works on proper mp4 videos, not with fragments (m4s). I cannot glue the wanted m4s tiles back into a bigger one. However, the JavaScript player only takes fragments, meaning that I cannot just send continuous mp4 video.

  


  I have tried many things with fragments and full videos, but the only solution I found for this chicken-egg problem is to :

  


  

  1. Create video clips with smaller duration for each tile, at each bitrate.

  


  2. Glue the desired video clips tiles together with the FFMpeg command.

  


  3. Convert that glued video in one big fragment, and send it.

  


  


  However, I feel that it is not a clean method and I fear that I am missing something.
Is there a better solution that I am not aware of ?
Any idea suggestions are welcome.

  


  Thank you for your time.

  


• Output image twice in ffplay

  2 mai 2024, par Sun Tzun

  I need to show two similar images from a webcam side-by-side.
I can achieve this with this piped command, but it runs with a few seconds latency :

  


  ffmpeg -f dshow -i video="Logi C310 HD WebCam" -filter_complex "[0:v][0:v]hstack=inputs=2[v]" -map "[v]" -f matroska - | ffplay -


  


  I'd prefer to do the same in just ffplay.

  


  I'd also like to have the resulting image cropped to a square size or to the size of the original image (like if the picture from webcam gets divided by half and the halfs took each other places).

  


  What I need in this end is a tool that helps to align the camera parallelly to the captured object, so any wrong angle would be perfectly visible on video if two opposite sides of the picture are placed alongside.

  


  PS. My platform is Windows 11.