Hi Tim, All
so I'm back with an update.
First of all my comparison was not pure j2k vs dirac. Instead I took
j2k spec as base (but I didnt reuse their codebase), replaced the
qcoder with real arithmetic and added some lightweight compression
optimizations. The testing was made so far on a single file CITY
(http://media.xiph.org/video/derf/) which was converted from 4CIF to
512x512 (limitation of my implementation so far) in intra-only mode.
The result is: my codec compresses about 2,5% better. My calculations
show that this is the job of my optimizations, so I think it would be
quite realistic to say that j2k + arith does same as dirac.
Timing: on my dell studio 1749 my codec spent 1.97701 sec/frame in
average while native dirac spent 1.7084 sec/frame as reported by
EncodeDirac.exe.
It would be interesting to know how much Schrodinger is faster
(probably I will try it if it compiles on Windows) because while my
current implementation is just horrible I noticed that about 40% of
time is consumed by arithmetic coder code which is more or less
optimal...
BR,
Vladimir
--- Исходное сообщение ---
От кого: "Tim Borer" <tim.bo...@bbc.co.uk>
Кому: javasharp...@ukr.net, schrodinger-devel@lists.sourceforge.net
Дата: 1 ноября 2010, 12:30:40
Тема: Re: [Schrodinger-devel] dirac vs jpeg2k entropy modelling
Hi,
I would be interested to hear the results of experiments comparing
Dirac Pro and J2K. I assume you will be comparing Intra Dirac,
i.e. Dirac Pro/SMPTE VC-2, with J2K?
As you say Dirac Pro's entropy modelling is quite different to
J2K. I doubt it yields a higher compression ratio - the objective
with Dirac Pro was to produce a simpler, less complex solution
rather than a higher compression ratio. However I would add that,
because Dirac Pro takes care to preserve the DC precision it can
have fewer visual artefacts than J2k (even if PSNR figures don't
show this). I say this because we have noticed that J2K, used for
video, can produce an annoying DC flicker even at quite low
compression ratios.
I did some (unpublished) experiments a few years ago comparing J2K
and Dirac Pro. With arithmetic coding enabled the compression
efficiency (for lossless coding) was almost identical for the two
codecs (some images were 1 or 2% better with one codec and the
other way round for other images). Interestingly when arithmetic
coding was disabled, so that entropy coding was only the fixed
variable length coding (interleaved exp-Golomb), the compression
efficiency of Dirac Pro only dropped by 20%. That is for lossless
coding arithmetic coding only appeared to give a compression gain
of 1.25. Given that arithmetic coding adds significantly to the
complexity of the codec some applications may prefer the option of
not using arithmetic coding in Dirac Pro for lower complexity at
the expense of slightly reduced compression.
At the moment there are low delay profile hardware implementations
of Dirac Pro/SMPTE VC-2 available from Numedia Technology and from
Westwood Rock. Software implementations are available in the Open
Source Schroedinger software and from Entropy Wave.
Regards
Tim
Dr Tim Borer
Lead Engineer
BBC Research & Development
BBC Centre House
56 Wood Lane
London W12 7SB
At 20:46 30/10/2010, javasharp...@ukr.net wrote:
Hi,
I want to do some experiments with dirac and j2k and I have a
question. As I see dirac's entropy modelling is totally
different from that of j2k. It would be interesting to know
the rationale behind it. In particular if the j2k's entropy
modelling (together with the dirac's arithmetic encoder which
as I understans is more effective than j2k's quasi-arithmetic
encoder) what size decrease can be expected?
Thanks in advance,
Vladimir
------------------------------------------------------------------------------
Nokia and AT&T present the 2010 Calling All Innovators-North
America contest
Create new apps & games for the Nokia N8 for consumers in
U.S. and Canada
$10 million total in prizes - $4M cash, 500 devices, nearly
$6M in marketing
Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to
Ovi Store
http://p.sf.net/sfu/nokia-dev2dev
_______________________________________________
Schrodinger-devel mailing list
Schrodinger-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/schrodinger-devel
No virus found in this incoming message.
Checked by AVG - www.avg.com
Version: 9.0.864 / Virus Database: 271.1.1/3225 - Release
Date: 10/28/10 18:34:00
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Schrodinger-devel mailing list
Schrodinger-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/schrodinger-devel
