[XviD-devel] Re: PSNR encoding

[Paul-Kenji Cahier]

I certainly think that this method would require a lot more cpu time but
that it
could be made so that you can choose a metric
This could make some really nice spatial constant quality because the actual
constant
quality-quantizer modes sometimes give some very variable results on some
encodes
Nandub had anti-shit which was kinda like that, and it prooved very useful
for it
Sure it would even better if you had the choice between some metrics
Sometimes cpu time doesnt matter, and you want a constant spatial quality



Date: Mon, 7 Apr 2003 17:26:41 +0000
From: elcabesa <elcabesa at inwind.it>
To: xvid-devel at xvid.org
Subject: [XviD-devel] PSNR encoding
Message-ID: <200304071726.41357.elcabesa at inwind.it>
Content-Type: text/plain;
  charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: list
Reply-To: xvid-devel at xvid.org
Message: 4

my brain has a very strange idea , very similar to fixed quant but slower
than
it=)

do you think it's possible setting a target PSNR and then choose a quantizer
for each frame to match this PSNR throug all the movie?

i think for each frame you should try encong it more than one time until
psnr
reach target, maybe this code can be reused also for better "quality
measure"
than PSNR, simply changing quality metric

this could be a tipical encoding ofr a frame x

try encoding whit q=4 psnr not meet
try encoding with q=3 psnr not met
try encoding with q=2 psnr meet
write only last stream=)

i don't know how actually xvid encode frames and if this can be applyed to
bframes too, or maybe it can  =)
maybe you would require only quant recalculation not motion estimation.

maybe i'm totally a MAD man who totally loose his small brain capacity=)
****************************************