[x264-devel] Re: some questions about Trellis RD

[Loren Merritt]

On Wed, 9 Nov 2005, lurui wrote:

> Hello:
> In the nearly added Trellis RD code,I can not understand below:
> 1.What is the lambda2_tab[] means?

lambda2_tab[] is simply an exponential sequence. I plan to merge it with 
the multiplications in x264_quant_4x4_trellis() to make it really the 
lambda values.

> Why we multipied the normal lambda by 1024?
> 2.Why the cost is
>    n.score += f8_bits * i_lambda2 >> ( CABAC_SIZE_BITS - LAMBDA_BITS );?
> I think CABAC_SIZE_BITS is needed for f8_bits, but I can't understand where the LAMBDA_BITS come from?

I needed some extra precision, and the various constants allow me to 
configure the precision in one place.

--Loren Merritt

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html