[x264-devel] Re: Patch for better RDO

[Loren Merritt]

On Fri, 15 Jul 2005, Robert Swain wrote:
> On 7/15/05, lurui <lurui.freesky at gmail.com> wrote:
>> Dear All, I think there is a bug in file ananlyse.c that result "RDO 
>> with --subme 6 is not really better than --subme 5". There shoule be a 
>> x264_analyse_update_cache() statement before x264_rd_cost_mb() statemet 
>> at line 1607 in analyse.c.
>>
>> Attachment is a patch file for it.
>>
>> LuRui

applied something similar (which also slightly helps with -m 5).

> I tested the patch and it gains 0.02dB overall on the short test clip
> used so it seems OK for me. I must say that I haven't encountered an
> occasion where -m 6 is not as good as -m 5 yet.
>
> Regards,
> Robert Swain

The occasion noted was paris_cif with -A all.

before: x264 paris_cif.yuv --frames 400 -A all -m 5
x264 [info]: PSNR Mean Y:37.53 U:40.49 V:40.60 Avg:38.33 Global:38.30 kb/s:482.3

before: x264 paris_cif.yuv --frames 400 -A all -m 6
x264 [info]: PSNR Mean Y:37.52 U:40.52 V:40.61 Avg:38.32 Global:38.30 kb/s:491.4

after: x264 paris_cif.yuv --frames 400 -A all -m 5
x264 [info]: PSNR Mean Y:37.55 U:40.51 V:40.61 Avg:38.34 Global:38.32 kb/s:481.3

after: x264 paris_cif.yuv --frames 400 -A all -m 6
x264 [info]: PSNR Mean Y:37.55 U:40.53 V:40.59 Avg:38.34 Global:38.32 kb/s:481.4

Ok, so RDO still doesn't help on that clip, but it's now no worse than 
non-RDO.

--Loren Merritt

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