[FFmpeg-devel] JPEG2000 decoder

Michael Niedermayer michaelni at gmx.at
Tue Aug 2 00:46:31 CEST 2011


On Mon, Aug 01, 2011 at 12:56:07PM -0700, Rukhsana Ruby wrote:
> Hi Michael,
> 
> On Wed, Jul 27, 2011 at 4:45 PM, Rukhsana Ruby <rukhsana.afroz at gmail.com>wrote:
> 
> > Hi Michael,
> >
> > Will you be able to be online again on IRC? I truly appreciate your help.
> >
> > Thanks
> >
> >
> > On Wed, Jul 20, 2011 at 5:54 PM, Rukhsana Ruby <rukhsana.afroz at gmail.com>wrote:
> >
> >> Hi Michael,
> >>
> >> I do not ask you to come online for hours, just for a few minutes. I found
> >> arithmetic encoding is working for the current decode, but after decoding
> >> component data differs between ours and jasper's. Its may be for our data
> >> representation. If you could come online, I could show you data very easily
> >> and this way my time would be saved a way more. In fact, I am stuck in this
> >> position. I truly appreciate your help.
> >>
> >> Thanks
> >>
> >>
> >> On Tue, Jul 19, 2011 at 1:25 PM, Rukhsana Ruby <rukhsana.afroz at gmail.com>wrote:
> >>
> >>> Hi Michael,
> >>>
> >>> Could you please be online on IRC for a while?
> >>>
> >>> Thanks
> >>>
> >>> On Fri, Jul 15, 2011 at 6:14 PM, Michael Niedermayer <michaelni at gmx.at>wrote:
> >>>
> >>>> On Sun, Jul 10, 2011 at 08:46:04AM -0700, rukhsana afroz wrote:
> >>>> > Hi Michael,
> >>>> >
> >>>> > On Fri, Jul 8, 2011 at 4:57 AM, rukhsana afroz <
> >>>> rukhsana.afroz at gmail.com>wrote:
> >>>> >
> >>>> > > Hi Michael,
> >>>> > >
> >>>> > >
> >>>> > > On Thu, Jul 7, 2011 at 8:09 AM, rukhsana afroz <
> >>>> rukhsana.afroz at gmail.com>wrote:
> >>>> > >
> >>>> > >> Hi Michael,
> >>>> > >>
> >>>> > >> Please respond to me whenever you see this message. I truly
> >>>> appreciate
> >>>> > >> your help. I will not take much of your time.
> >>>> > >>
> >>>> > >> Thanks
> >>>> > >>
> >>>> > >>
> >>>> > >> On Thu, Jul 7, 2011 at 7:19 AM, rukhsana afroz <
> >>>> rukhsana.afroz at gmail.com>wrote:
> >>>> > >>
> >>>> > >>> Hi Michael,
> >>>> > >>>
> >>>> > >>>
> >>>> > >>> On Thu, Jul 7, 2011 at 7:11 AM, rukhsana afroz <
> >>>> rukhsana.afroz at gmail.com
> >>>> > >>> > wrote:
> >>>> > >>>
> >>>> > >>>> Hi Michael,
> >>>> > >>>>
> >>>> > >>>> It would be great if you could manage 10/20 minutes for me. In
> >>>> your
> >>>> > >>>> convenient time, I will be on IRC.
> >>>> > >>>>
> >>>> > >>>> Thanks
> >>>> > >>>>
> >>>> > >>>>
> >>>> > >>>>
> >>>> > >>>> On Wed, Jul 6, 2011 at 6:43 PM, rukhsana afroz <
> >>>> > >>>> rukhsana.afroz at gmail.com> wrote:
> >>>> > >>>>
> >>>> > >>>>> Hi Michael,
> >>>> > >>>>>
> >>>> > >>>>> Are you in vacation? I dont see you online.  I urgently need
> >>>> your help
> >>>> > >>>>> on one thing. If you could tell me your convenient time, I can
> >>>> be online on
> >>>> > >>>>> IRC during that time.
> >>>> > >>>>>
> >>>> > >>>>>
> >>>> > >>>>> Thanks
> >>>> > >>>>>
> >>>> > >>>>> --
> >>>> > >>>>> Rukhsana Ruby
> >>>> > >>>>> Phd Student
> >>>> > >>>>> Department of Electrical & Computer Engineering
> >>>> > >>>>> The University of British Columbia
> >>>> > >>>>> ============================
> >>>> > >>>>>
> >>>> > >>>>>
> >>>> > >>>>>
> >>>> > >>>>
> >>>> > >>>>
> >>>> > >>>> --
> >>>> > >>>>
> >>>> > >>>>
> >>>> > >>> The current bug is due to the incorrect calculation of band
> >>>> coordiantes.
> >>>> > >>> This is due to the following variables in SIZ marker:
> >>>> > >>>
> >>>> > >>> XRsizi:Horizontal separation of a sample of ith component with
> >>>> respect to
> >>>> > >>> the reference grid. There is one
> >>>> > >>>       occurrence of this parameter for each component.
> >>>> > >>> YRsizi:Vertical separation of a sample of ith component with
> >>>> respect to
> >>>> > >>> the reference grid. There is one
> >>>> > >>>       occurrence of this parameter for each component.
> >>>> > >>>
> >>>> > >>> XRsizi: 2, YRsizi: 1 for the file p1_01.j2k. I am not very sure
> >>>> how to
> >>>> > >>> use these variables correctly to calculate band coordinate. I
> >>>> truly
> >>>> > >>> appreciate your help.
> >>>> > >>>
> >>>> > >>>
> >>>> > >>> Thanks
> >>>> > >>>
> >>>> > >>>
> >>>> > >>>
> >>>> > > I have fixed the bug in the calculation of band coordinates. Now, I
> >>>> have
> >>>> > > found new bug. I need your help on this. If you could reply, that
> >>>> would be
> >>>> > > very helpful for me.
> >>>> >
> >>>> >
> >>>> > I have fixed the bug on calculating the band coordinates and now
> >>>> arithmetic
> >>>> > encoding is working. I have one question, after dequantization before
> >>>> > inverse wavelet transform, should the component data of our decoder
> >>>> match
> >>>> > with jasper's decoder. I am finding mismatch between both data.
> >>>>
> >>>> yes they should match.
> >>>> It may be that the way things are stored differs, like sign/abs vs
> >>>> twos completent or things like that but the values should (ignoring
> >>>> how they are represented) match
> >>>>
> >>>>
> >>>> > Also, I am
> >>>> > getting memory leak while printing component data. My code for
> >>>> printing
> >>>> > component data is below:
> >>>> >
> >>>> > static void printcomp(J2kDecoderContext *s, J2kComponent *comp) {
> >>>> >     int i, j;
> >>>> >
> >>>> >     av_log(s->avctx, AV_LOG_INFO, "y: %d\n", comp->coord[1][1] -
> >>>> > comp->coord[1][0]);
> >>>> >     for( i = 0; i < comp->coord[1][1] - comp->coord[1][0]; i++){
> >>>> >        for( j = 0; j < comp->coord[0][1] - comp->coord[0][0]; j++)
> >>>> >      av_log(s->avctx, AV_LOG_INFO, "0x%x ", *(comp->data + j));
> >>>> >
> >>>>
> >>>> >        if (i < comp->coord[1][1] - comp->coord[1][0] - 1)
> >>>> >             comp->data = comp->data + comp->coord[0][1] -
> >>>> comp->coord[0][0];
> >>>>
> >>>> that does more than just printing
> >>>> and to debug memleaks try valgrind, its quite powerfull and will
> >>>> pinpoint exactly where the leak is
> >>>>
> >>>> [...]
> >>>> --
> >>>>
> >>>
> 
> I have put two pastebins (with partial data) which are the component data
> before dequantization. These data is for the file which can be decoded
> correctly. I think, data representation is way different between ours and
> jaspers and therefore we see the difference in each and every byte of
> component data. I want to modify ffmpeg's code so that data representation
> looks similar to jasper. Could you please tell me how i can do it?
> Otherwise, its impossible to debug for the file which cannot be decoded at
> all.
> 
> ffmpeg:
> http://pastebin.com/kAMmZh30
> 
> jasper:
> http://pastebin.com/mya6RxNS
> 
> Please get back to me as early as possible. I want to finish this work as
> soon as possible. And sorry, I was busy with my school work last few days
> and didnt work on it. Also, I am on IRC.

by looking at it you can see that

-2 -> -5
-1 -> -3
0 -> 0
1 -> 3
2 -> 5
3 -> 7
6 -> 13

it should be easy to see from this how to map between the 2
representations

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I have never wished to cater to the crowd; for what I know they do not
approve, and what they approve I do not know. -- Epicurus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20110802/c9cc4f5a/attachment.asc>


More information about the ffmpeg-devel mailing list