[XviD-devel] zig-zag

[Christoph Lampert]

On Wed, 12 Mar 2003, Christoph Lampert wrote:
> But maybe the peaks are the reason why HQAC-PRED works. Btw. that just
> checks for 1 directions (which is write to acpred_directions[block] in
> predict_acdc()). Shouldn't somehow both be checked? 

Sorry, no. If I see it right, then AC prediction can just be switched on
or off, the direction is determined by DC prediction and that the encoder
can't decide on! It's calculated from neighbouring blocks DCT gradient. 

This sound horrible... is that right?

gruel