Re: YV12 decoding

>From: Rrrough Spirit <[email protected]>
>(which admittedly isn't THAT important).
>sh0dan stated, that the needed functions are basically
>already there, so my question is, would it be
>painlessly possible to modify xvid that way, that it
>always decodes (raw) YV12 streams and reports it to
>virtualdub "upon request" ? Or maybe at least include
>this hack into CVS ? I'd love to be independant from
>DivX5, and still have a mighty fast compression
>without any color conversions.

hi,

ive commited the yv12->yv12 memcpy hack into cvs.

does anyone think xvid should provide application access all our
colorspace conversion functionality via the api? OR, should we only
provide yv12->rbg as a quick fix until virtualdub supports yv12??

i can see five levels upon which color conversion can be implemented:

1. basic hack. provide a simple YV12->xxx
==========================================
       opt = XVID_INIT_YV12TOXXX
       typedef struct
       {
           int width;
           int height;
           DEC_PICTURE * input;  // YV12 input image
           int colorspace;     // output colorspace
           void * output;      // output buffer
           void * stride;      // output stride
       } XVID_INIT_YV12PARAM;

2. add support for all conversions
==================================
      opt = XVID_INIT_CONVERT
      typedef struct
      {
          int colorspace;
          int width, height;
          void * y_ptr;
          void * u_ptr;
          void * v_ptr;
          int y_stride;
          int uv_stride;
      } XVID_IMAGE;        // passed as param1 and param2

      param1 = destination,
      param2 = source
      when colorspace is packed, only y_ptr and y_stride as used.
      return fail if the width/height do not match.

3.  modify encoder/decoder to use XVID_IMAGE
============================================
the XVID_IMAGE structure neatly describes all the properties of
an image, so we could use this within the encoder and decoder.

4. remove image conversions from encoder/decoder
================================================

* rename xvid_init to xvid_global.

* remove colorspace stuff from the encoder and decoder, and place it
  into the global function. (XVID_INIT_CONVERT becomes XVID_GLO_CONVERT)

* upon creation, the encoder and decoder return thier internal
  buffers. colorspace handling if left upto the application.
  (much like mpeg2dec).

e.g. encoding:

   xvid_global(0, XVID_GLO_INIT, xinit);       // init xvid
   xvid_encode(0, XVID_ENC_CREATE, xparam);   // create encoder
   while( ...frames...) {
      xvid_global(0, XVID_GLO_CONVERT, my_src_bmp, xparam.image);
      xvid_encoder(xparam.handle, XVID_ENC_ENCODE);
   }
   xvid_encoder(xparam.handle.

e.g. decoding:

   xvid_global(0, XVID_GLO_INIT, xinit);       // init xvid
   xvid_decode(0, XVID_DEC_CREATE, xparam);   // create encoder
   while( ...frames...) {
      xvid_encoder(xparam.handle, XVID_DEC_ENCODE);
      xvid_global(0, XVID_GLO_CONVERT, xparam.image, my_dst_bmp);
   }
   xvid_encoder(xparam.handle.

5. write a seperate libcolorspace, or libxvidcolorspace or something
====================================================================
?!

iam happy with number 2 (or maybe 3). any other opinions?

-- pete

_________________________________________________________________
Get faster connections -- switch to MSN Internet Access! 
http://resourcecenter.msn.com/access/plans/default.asp