[XviD-devel] Reduced-Resolution P-frame

[skal]

	Hi Pete and all,

On Fri, 2002-11-29 at 11:44, peter ross wrote:
> hi skal,
> 
> i like too, but there more important stuff to do (ie. APIv3)
> a month back, i added very rough reduced resolution i-vop decoding
> 8x8 blocks are decoded, but are not rescaled to 16x16
> 
> if you have written the scaling/filtering code, then it should not
> be much work to implement it. but like you say, there's a lot of
> auxillery testing necessary to ensure reduced res works with
> interpolation, bframes(?), qpel(?), etc.

	The good news: reduced-VOP is not defined for B-VOP :)
	The bad news: i would have bet B-VOP were the best
	place to take advantage of reduced-resolution, between
	two full-resolution P-VOP.

	Besides, I found reduced-VOP integrates rather seamlessly
	with qpel, field-prediction,... except for the few
	pitfalls I mentioned. There might be problems with smarter
	ME algorithm than the rudimentary one I used. In particular,
	scaling down 2x2 macroblocks bearing various decisions 
	into a single one merged to apply reduced-resolution coding
	on might reveal being tricky.
(nightmare example: how should we merge:
 [SKIP     ][INTER4V         ]
 [INTRA/DCT][FIELD-PRED INTER] into a single macroblock? Not to
mention CBP...:)

	Anyway, gimme 1 day (maybe less) to put the filtering code
	into 'xvid-like' shape and I'll post it here...

	later,
		Skal