[x264-devel] Re: Test results of trellis,brdo,and mixed_refs

[Guillaume POIRIER]

Hi,

On 12/15/05, Tuukka Toivonen <tuukkat at ee.oulu.fi> wrote:
> I noticed that there were a bunch of new options in x264.
> I made a quick testing of them, using the methods I think
> I have described sometime: ie. fixed QP values at several
> values, around 12 640x480 sequences, 250 frames or so.
>
> The results are presented as percentage of the original
> bitrate at the same quality (PSNR), smaller = better.
>
> Looks like every one of the new options helps significantly. Great! In
> total by using all of the options, around 7% of bitrate reduction is
> obtained (at the same PSNR). Especially trellis makes nice impact.

I observed the same thing too.


> I didn't put the EPS-graphs & timings online, but if someone really wants
> them I can provide those too.
>
> Exact options and results are below (for mencoder):
>
> x264best2: baseline
> "$OPTS:nomixed_refs:nobrdo:trellis=0:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"
>
> x264best2mr: mixed references
> "$OPTS:mixed_refs:nobrdo:trellis=0:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"
>
> x264best2brd: brdo (B-frame MB-type RD-optimization)
> "$OPTS:nomixed_refs:brdo:trellis=0:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"
>
> x264best2tr1: trellis=1
> "$OPTS:nomixed_refs:nobrdo:trellis=1:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"
>
> x264best2tr2: trellis=2
> "$OPTS:nomixed_refs:nobrdo:trellis=2:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"
>
> x264best2mrbrdtr2: all new options enabled
> "$OPTS:mixed_refs:brdo:trellis=2:frameref=15:nob_adapt:b_bias=0:bframes=3:b_pyramid:ip_factor=1.0:pb_factor=1.5:direct_pred=1:noweight_b:i4x4:i8x8:b8x8mv:8x8mv:no4x4mv:8x8dct:me=3:me_range=24:subq=6:nochroma_me:chroma_qp_offset=8"

Is there a reason why you added these specific options? Is that to
compare x264 VS reference H.264 code, or did you notice that this
setting looked better for your needs? I've read several times that
adaptive bframes, weighted bframes and pyramid were pretty much always
improving quality.
As far as me_range is concerned, I've tested 2 encodes with uneven
multi-hexagon search and me_range=16 (default) and me_range=64 (the
maxme_range possible). From the top of my head, it slowed down
encoding by a factor of 2 while adding just 0.01dB PSNR. Now, that was
just a single test on asource that maybe could not benefit from a big
me_range... and you may have found sources that do benefit from a big
me_range...

Guillaume

--
MPlayer's doc is offline. Find some fresh one here:
http://tuxrip.free.fr//MPlayer-DOCS-HTML/en/
http://tuxrip.free.fr//MPlayer-DOCS-HTML/fr/

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