Recherche avancée

Sur d’autres sites (12610)

• intreadwrite : Add intermediate variables in the byteswise AV_W*() macros

  1er août 2016, par Martin Storsjö

  intreadwrite : Add intermediate variables in the byteswise AV_W*() macros
  This avoids issues with expanding the argument multiple times,
  
  and makes sure that it is of the right type for the following shifts.
  Even if the caller of a macro could be expected not to pass parameters
  
  that have side effects if expanded multiple times, these fallback
  
  codepaths are rarely, if ever, tested, so it is expected that such
  
  issues can arise.
  Thefore, for safety, make sure the fallback codepaths only expand
  
  the arguments once.
  Signed-off-by : Martin Storsjö <martin@martin.st>
  

  • [DBH] libavutil/intreadwrite.h

• AVFrame in AV_PIX_FMT_YUV420P format to H*W*3 data buffer conversion using sws_scale not accurate ?

  13 février 2024, par user3133806

  I have an AVFrame that is in YUV420 format and I am trying to convert to packed RGB in a data buffer :

  


    // Allocate enough for 8 bits per color in RGB.&#xA;  // Note that I want packed, not planar. i.e. RGBRGBRGB...&#xA;  // Each pixel is 3 bytes of RGB in row-major order&#xA;  buffer = new unsigned char[frame->height * frame->width * 3];&#xA;&#xA;  enum AVPixelFormat frameFormat =&#xA;      static_cast<enum avpixelformat="avpixelformat">(frame->format);&#xA;&#xA;  // Make sure the input is in the expected format.&#xA;  CHECK_EQ(frameFormat, AV_PIX_FMT_YUV420P);&#xA;&#xA;  SwsContext* swsContext = sws_getContext(&#xA;      frame->width,&#xA;      frame->height,&#xA;      frameFormat,&#xA;      frame->width,&#xA;      frame->height,&#xA;      AV_PIX_FMT_RGB24,&#xA;&#xA;      // I played around with these flags but could not get an accurate image&#xA;      SWS_BICUBIC | SWS_ACCURATE_RND,&#xA;      nullptr,&#xA;      nullptr,&#xA;      nullptr);&#xA;&#xA;  // My understanding is that because I am using packed format, not planar&#xA;  // I only need to fill in the first pointer.&#xA;  unsigned char* out_planes[4];&#xA;  out_planes[0] = buffer;&#xA;  out_planes[1] = nullptr;&#xA;  out_planes[2] = nullptr;&#xA;  out_planes[3] = nullptr;&#xA;&#xA;  sws_scale(&#xA;      swsContext,&#xA;      frame->data,&#xA;      frame->linesize,&#xA;      0,&#xA;      frame->height,&#xA;      &amp;out_planes,&#xA;&#xA;      // My understanding is that the stride of each plane is frame->width * 3&#xA;      (const int[4]){frame->width * 3, 0, 0, 0});&#xA;  sws_freeContext(swsContext);&#xA;&#xA;</enum>

  


  This code does produce an image but the image appears to be wrong.

  


  On a 300x200 video, the right few columns are wrong.

  


  On a bigger resolution video the image seems correct, but on a smaller video it is not :

  


  

  


  

  


  EDIT : It appears this is a bug in sws_scale (bug or expected feature). Turning off vector instructions causes it to work as expected :

  


  // Adding this line before sws_scale causes the images to be bit exact.
av_set_cpu_flags_mask(0);


  


  I got that idea from this answer on SO : Turn off sw_scale conversion to planar YUV 32 byte alignment requirements

  


• Révision 18371 : Report de r18366 :

  22 août 2011, par cedric -

  Manque dans l’analyseur de DTD : il ne verifie pas qu’un attribut de type NMTOKEN ou NMTOKENS ne contient que des noms. Utile pour certains attributs des balises d’en-tete dans la DTD XHTML Basic 1.1, mais curieusement pas pour l’attribut "class" qui est clairement un NMTOKENS pourtant. (...)