[x264-devel] PSNR vs the distance of reference frame

[john sisi]

Thanks a lot Dylan,

so you are saying that if I use the ASNR instead of PSNR, then the
curve (ASNR vs distance) will be decreasing? Thanks again,
John

On Tue, Jun 29, 2010 at 5:35 PM, Dylan Klebesadel
<dylanklebesadel at gmail.com> wrote:
> The reason for the increace in PSNR is the P.
> Think of what the P stands for.
> Your ASNR would be the same or lower, however.
> Try doing this with the increace to 14 and collecting the ASNR and PSNR
> going downwards and plot a graph.
> You'll see what I mean :)
> Peak Signal to Noise Ratio
> Average Signal to Noise Ratio
> The latter is more important for compression and the former is more
> important for finding how to fix the ladder (when what is going on when it
> reaches peak)
> Logs please?
>
> On Tue, Jun 29, 2010 at 5:25 PM, john sisi <sisi.john1 at gmail.com> wrote:
>>
>> Dear All,
>>
>> I performed an experiment using x264 to find the effect of using
>> different reference frames on the PSNR performance. I got a somehow
>> weird result: it seems that the PSNR goes up (by about 0.2dB on
>> Foreman CIF @256kbps) as the distance to the reference frame is
>> increased (from 3 to 15).
>>
>> I set the number of reference frame to 15 using "--ref 15" and I
>> changed the code such that I was able to select only one of these 15
>> reference frames for encoding the frames. For example, in
>> "x264_mb_analyse_inter_p16x16" I set "i_ref" to 5 just before
>> "REF_COST( 0, i_ref )" to use only the 6th reference frame. I applied
>> the same change on the other relevant functions like
>> "x264_mb_analyse_inter_p16x8".
>>
>> My expectection is that the PSNR values must go down as we use farther
>> reference frames but is this result reasonable?  Thanks,
>> John
>> _______________________________________________
>> x264-devel mailing list
>> x264-devel at videolan.org
>> http://mailman.videolan.org/listinfo/x264-devel
>
>
> _______________________________________________
> x264-devel mailing list
> x264-devel at videolan.org
> http://mailman.videolan.org/listinfo/x264-devel
>
>