[x264-devel] Question about SAD and SATD in intra mode decision algorithm

[Jason Garrett-Glaser]

On Sun, Mar 14, 2010 at 7:51 PM, kenter <kenter83 at gmail.com> wrote:
> Hi,
>      I currently study on intra mode decision algorithm of x264, I found
> that there are 2 cost computation func: SAD and SATD. I know what they
> means, while I puzzled about their compression performance. I want to
> implement my project by using SAD and I test some sequencies: Akiyo(QCIF),
> Bridge-close(CIF) and forman(CIF) at low bit-rate condition. It seems there
> are not much PSNR drop.Now, I want to ask: why not SAD as default? in which
> condition the SATD will suprior to SAD?
>     Thanks a lot!

2-pass, intra-only, foreman CIF 2000kbps, --tune psnr:

SAD mode decision:

x264 [info]: PSNR Mean Y:36.250 U:40.452 V:42.106 Avg:37.355
Global:36.983 kb/s: 2014.53

SATD mode decision:

x264 [info]: PSNR Mean Y:36.493 U:40.554 V:42.192 Avg:37.576
Global:37.154 kb/s: 2016.10

Looks quite a bit better to me.

Dark Shikari