[MPlayer-dev-eng] Recommend error for bad dimensions

[D Richard Felker III]

On Mon, Mar 01, 2004 at 11:02:25PM -0600, Zoltan Hidvegi wrote:
> > And as for v4mv, I don't think it will help at all. You still have to
> > encode full macroblocks (16x16). Aligning at multiple-of-8 is better
> > than no alignment at all, though.
> 
> I thought that the DCT coding is done for 8x8 blocks.  So if your
> dimensions are a multiple of 8, then the luma coding will be fine,
> only the chroma part can have a partial block.

Yep. That's why 8 is better than no alignment at all. But you still
waste bits on the partial chroma block and the overhead for the
macroblock only gives you 2 luma blocks rather than 4.

> But when motion vectors originate from outside of the picture, a
> portion of the edge blocks will not have any prediction information
> even if the dimensions are multiple of 16.  Do B-frames have
> predictors from both time directions, so they can be better at the
> edges?

Yes, I would expect that B frames might help the situaion, but I've
never seen any empirical comparison.

Rich