[MEncoder-users] "High Quality" H.264 at low bit rates ;)

[Grozdan Nikolov (openSUSE Linux)]

On Monday 03 March 2008 15:42, Gunter Ohrner wrote:
> Grozdan Nikolov (openSUSE Linux) wrote:
> >> What PSNR and SSIM results should I be able to expect with 2 MBit?
> >> Can anyone of you provide some values to compare?
> >
> > PSNR/SSIM highly depends on the source/bitrate/resolution so you can't
> > compare a source that easily with two or more different sources
>
> Ok.
>
> > @ cropped standard resolution and bitrate of 1020 kbps, i usually get
> > SSIM between 0.976 and 0.986 but again this highly depends on the source
> > and the x264 settings you use
>
> For my test clip 720x544 @ 2000kbps (cropped to avoid black borders) I get,
> using a current libx264 including all three patches you suggested, SSIM
> 0.9745643. (3-pass encoding)
>
> mencoder parameters were
>
> nointerlaced:subq=7:partitions=all:8x8dct:me=umh:frameref=5:bframes=3:b_pyr
>amid:weight_b:me_prepass:psnr:ssim

partitions=all is basically useless and only slows down the encoding by 
activating the p4x4 partions which in reality provides less than 0.2% in 
compression but slows down encoding a lot. So better use 
partitions=p8x8,b8x8,i8x8,i4x4

also if you're using b-frames then enable brdo and bime options and set 
direct_pred=auto ... enable trellis=1 and cabac too

there's also a quantizer noise shaping patch but it's really really slow and I 
don't know how compatible is with the current x264 code base

>
> So this still seems to be somewhat low in comparison, doesn't it? I needed
> 3000kbps to push SSIM > 0.98. At 3000kbps, the video looks "acceptable" for
> me. With 1000kbps I merely get 0.9559612...
>
> Is there anything obviously wrong with my settings? I verify I'm not doing
> something obviously wrong I also tested the h264enc script which produces
> comparable results, however - at least it does not achieve significantly
> better quality.
>
> Greetings,
>
>   Gunter
>
> _______________________________________________
> MEncoder-users mailing list
> MEncoder-users at mplayerhq.hu
> https://lists.mplayerhq.hu/mailman/listinfo/mencoder-users

-- 
Regards