[XviD-devel] Inquiry Skal About Decoding field predicted MacroBlock

skal skal at planet-d.net
Wed Dec 15 19:56:18 CET 2004


	Hi Mayfig,

On Wed, 2004-12-15 at 10:02, mayfig wrote:
> Hi,
>     
>      When Xvid decodes an  interlaced sequence which includes field-predicted
> MB, it does not function well ,for it has problem on the motion compensation in
> this case, i am trying to correct it, although has got some improved result,
> but still 
> not satisfactory.
> 
>     I noticed that Skal has good implementation in his "sklmp4-0.23" for the
> special case, so could Skal reply some of my inquiry?
> 
> 1) in the interlaced mode, the decoded reference MB:( dec->refn[ref].y
> ,dec->refn[ref].u,dec->refn[ref].v,) and the decoded current MB
> (dec->cur.y,dec->cur.u,dec->cur.v) , are organized in the mode of Frame
> Picture, i.e. , the two fields of the frame are interleaved, Am I right?

	yep. (=> to access fields, you just have to double the stride)

> 
> 3) With respect to Subclause 6.3.6.3, also taking the
> forward_top_field_reference 
> and forward_bottom_field_reference flag into account. And I think the
> referenced field comes from the previous VOP, am i right?  

	Indeed, at the beginning of a VOP, you must re-use the field
	references (top bottom *and* forward/backward) of the last
	macroblock of the previous VOP. This is also important to store
	field reference in P-VOP, in case a field-predicted BDirect
	mode shows up in a subsequent B-VOP.

	hope it helps,
Skal




More information about the XviD-devel mailing list