[sldev] Meaning of RequestImage DiscardLevel field?

[Teravus Ovares]

Once again, I know this is speculation..    however I ran some tests
on this a few months back.   I attempted to send the discard levels as
individual images, and the first discard level requested came through,
however, subsequent 'images' sent representing the next level of
discard level caused it to corrupt the image client side.    This
leads me to believe that the discard levels represent byte offsets of
where the boundary between the packets that represent the discard
levels are..   as far as the server is concerned.   So, in other
words..   lets say you have 5 discard levels in an image.

You'd have to know this before hand in meta data or decode it on the fly.

The client asks for discard level 5...      We send it bytes 0-1000.
The client asks for discard level 3 of the same image..    we know it
has already received the discard level 5 bytes..     so we send bytes
1001-5000.

..   and we keep appending the rest of the file based on the packets
in the file

This implies that the packets are ordered in sequence from lowest at
the beginning to highest at the end.

Sometimes we get requests on the simulator for -1 before the rest of
the discard levels are requested.

Sometimes we get requests on the simulator for -1 when the third
discard level is requested..       It seems to have to do with
defining and keeping image priority.

On a Linden Simulator, it appears that there's an image priority queue
that it uses to feed the clients images at different discard levels
and keep track of the ones it has previously sent.   It also seems to
reorder them in the priority 'list' that the server keeps.

But!, it would be nice to get a LL Developer's view on what Discard
Level actually represents.

Best Regards

Teravus

On 9/24/08, Salvador Agati <agati at procele.com.br> wrote:
>
>
>
> ----- Original Message -----
> From: John Hurliman
> To: Carsten Juttner
> Cc: Second Life Developer Mailing List
> Sent: Sunday, September 21, 2008 1:05 AM
> Subject: Re: [sldev] Meaning of RequestImage DiscardLevel field?
>
>
> Thank you Carsten, that clarification is correct. For now I'm assuming that
> DiscardLevel 0 is a request for the full quality (zero discarded levels) and
> DiscardLevel 5 is requesting the lowest quality level. However, if you start
> with five quality layers and you discard five of them what are you left
> with? Is five actually being clamped to four? This still doesn't explain -1.
> If you send a RequestImage with Priority = 0.0 and DiscardLevel = -1 it will
> cancel the download, but texture requests often start with a positive
> priority and a DiscardLevel of -1. Is this a request for the header only, or
> is it just an uninitialized value in the client that implies zero or four?
>
> John
>
>
> On Fri, Sep 19, 2008 at 3:45 PM, Carsten Juttner <carjay at gmx.net> wrote:
> >
> >
> > Robin Cornelius wrote:
> >
> > John Hurliman wrote:

> >
> > I know that the DiscardLevel field in the RequestImage packet is used
> to
request different quality levels (not different texture sizes as
> the
protocol documentation states, SL uses LRCP ordered JPEG2000 files),
> but
i can't figure out what the values correspond to. In a typical
> texture
download I'll see values ranging from -1 to 5. Is -1 a special
> value? Is
there an upper limit? Does a larger number mean a lower quality
> layer?


> > The discard field is directly related to the request quality level as
you
> are already aware. It should be a simple calculation that the
discard is the
> number of powers of 2 to scale the image down by for the
reduced quality. So
> a discard of 0 is the complete image and a discard
of 1 is a 1/2 size, a
> discard of 2 is a 1/4 size etc.

> >
> > I think what John was actually referring to here is that quality layers
> and resolution levels are not the same. In JPEG2k all data is divided into
> packets for one tile. Each packet contains codestream data for one quality
> layer for one resolution for one component for one precinct (spatial
> partition). The order in which you transmit the packets decides what you get
> first and is signalled by the "progression order". LRCP
> (Layer-Resolution-Component-Position) is layer-centric
> meaning that one by one you get all the data for each quality layer, not for
> each resolution. That means you end up with the lowest quality layer for all
> resolutions first so it's really a quality-based progression, not a
> resolution-based.
> >
> > Think of the quality layers as going down the bits from MSB to LSB. First
> you get the rough information, then it progresses to the detailed
> information (actually for the wavelet coefficients).
> >
> > AFAIK, inside the viewer the progression order is not really cared about
> and the discard levels are merely referring to the dropped resolutions from
> the highest resolution.
> >
> > Also nobody seems to really know what is happening on the server side
> since we did ask a while ago but never met a Linden who was familiar with
> that area of the code so it's still a bit of a mystery.
> >
> > Regards,
> > Carsten
> >
> >
> >
> >
> > I'm not used to post here and I' ve got only these 2 messages above
> related to the subject. Indeed , english is not my native language and I am
> concious that I'm not an expert as you, in this list,  are.
> >
> > Despite of that, I decided to send this post because one of the first
> things I noted when I began to discover SL was exactly the way images were
> loaded in the viewer.
> >
> > It reminded me when I,  after the college, was a mastering student and had
> to read and discuss, in the class, the implementation of  an algoritm . I
> happened at about 20 years ago.
> >
> > In this book, it was presented an algoritm called "gross information
> first" and it used in the formula,  power of 2 numbers to calculate the
> pixel value of a group of pixels of a image.  So, the quality of the image
> generated with the algoritm was related to 2, 4, 8, 16, 32 or thinking as
> power of 2,  the values 0,1,2,3,4,5 .
> >
> > I was wondering if this is the case, where the viewer asks for the next
> quality layer after the previous layer loaded and -1 to ask for data without
> this type of filtering.
> >
> > In this situation, the image, in the server, would  be stored only with
> the highest quality and the server would  first "filter" the image, with the
> correct factor, and then,  send it to the viewer, progressively, as
> requested.
> >
> > If someone call me crazy , I won't deny.
> >
> > Regards,
> > Agathos Frascati SL
> > Salvador Agati RL
> >
> >
> >
> >
> >
>
> _______________________________________________
> Policies and (un)subscribe information available here:
> http://wiki.secondlife.com/wiki/SLDev
> Please read the policies before posting to keep unmoderated posting
> privileges
>