[XviD-devel] Mode decision

[skal]

	Hi,

	I've been conducting some tests on this old subject:

On Thu, 2003-02-13 at 14:52, Christoph Lampert wrote:
> Hi,
> 
> from the logfile which a modified version of syskins mode decision
> generates, it seems clear to me, that the decision INTER/INTER4V is the
> crucial part. In foreman and miss_america, INTRA blocks are hardly
> present, but still BITS-decision manages to significanty lower bitrate. 
> 
> I plotted the "sad16 / sad8"-values of those blocks who were classified 
> INTER using bitcount (green) and those who classified INTER4V (red).
> The blue line is approximately our current decision 
> method: sad16<sum_of_sad8 (the extra penalty is only 2*quant, it would 
> hardly be visible in the plot). 
> 
> What you can see is that we are _very_ conservative when it comes to
> choosing INTER4V. All "real" INTER are classified correctly, but a vast
> number of "real" INTER4V is classified as INTER. 
> 
> You can also see, that using current sad16/sad8, no good decision
> boundaries are possible at all, because the area completely overlap. 
> 
> What you can't see, is how much a misclassification of INTER->INTER4V
> or vice versa would cost in extra bits. 
> And you also cannot see that there are hardly any real INTER4V blocks
> with sad16 below 600. So maybe this would be a way to speed things up, 
> simply not doing inter4v for those.

	Even with a rather crude sad16-vs-sad8 final criterion for
	INTER/INTER4v decision, I've tried to guess whether a 16x16
	INTER block was worth a candidate for a sub 4V search (which
	takes times). I've tried some criterion based of local gradient
	and divergence (is the MV-field torn appart?), but it's not
	convincing. Actually, it seems that 4V works best at image's
	segmentation limits. So far, so good, I ended up with a sad
	-based criterion that is a good hint of whether going for a
	sub search might be rewarding: after a 16x16 regular search,
	I re-use the best MV found so far to evaluate sad8 of each of
	the four 8x8 sub blocks. If the maximum of these sad8 values 	is
greater than a fraction the sad16 for the full block
	(in practice 60% is a good compromise), then I go for a
	refined search...

	Any opinion?

	bye,
		Skal