[sldev] GSoC - mesostructures

[Ravish Mehra]

hi Suzy,

1) actually I am not using the 3D texture for doing any kind of volumetric
texture mapping ( texture slices ).  it just stores the distance map( which
is actually a data structure - kind of a 3D grid which stores at each
location, the distance to the nearest point on the surface). this distance
map is accessed at each step of sphere tracing to determine the next point
to go to.

2) the central library - functions for creating the data structures used by
these various techniques and for binding the shaders will be in C++. the
vertex and fragment shaders will be in glsl. the purpose is to provide a
complete support for doing per-pixel displacement mapping.

at present, none of my texture maps( 2D or 3D ) are floating points as 8 bit
accuracy is enough for now. but i have certainly worked with floating point
textures and it has some advantages in certain cases.

thanks for you interest.

best
ravish

On Sat, Jul 12, 2008 at 7:08 PM, Suzy Deffeyes <suzyq at pobox.com> wrote:

>
> Hi Ravish,
>
> Thanks for the explanation. This is exciting stuff.  Some questions:
>
> 1. I wanted to confirm, "3D texture" to me means volumetric texturing using
> texture slices. Is that what you are using in your sphere tracing?  If so I
> can certainly understand the need to stay at lower resolutions... :). Volume
> texturing chows on texture memory (and in SL's case network bandwidth) like
> there is no tomorrow.
>
> 2. The mesostructures wiki page says it is creating a C++ library to do
> these techniques. Wouldn't these be implemented instead as GLSL shaders?
> I'd also be curious if any of your maps are made up of floating point
> values.
>
> Again, great stuff, I look forward to seeing your work as it progresses.
> Suzy Deffeyes/Pixel Gausman
>
>
> On Sat, Jul 12, 2008 at 5:32 AM, Ravish Mehra <ravish.mehra07 at gmail.com>
> wrote:
>
>> hello guys,
>> my name is Ravish and i am currently implementing this mesostructure
>> rendering library under google summer of code.
>>
>> i know i write long answers :) , so bear with me. regarding the visual
>> difference between the three techniques -
>> 1)  sphere tracing uses a 3D texture known as distance map for finding the
>> ray-mesostructure intersection and the quality achieved by this technique is
>> highly dependent on the resolution of this distance map. currently its
>> 128x128x32. since its a 3D texture we cannot go to very high resolutions.
>> hence the images will be at that resolution and certain aliasing effects are
>> visible. i have marked such areas in attached image with yellow circles.
>>
>> 2) Relief mapping's rendering quality on the other hand, depends more on
>> the number of iterations we do ( linear + binary steps ). if we do enough
>> iterations, it gives amazing results. as can be seen that the aliasing
>> effects present in sphere tracing are not present in relief mapping ( height
>> field ). but if we do less number of iterations, we may get certain
>> artifacts due to incorrect intersections(see attached figure of relief
>> mapping). also since it uses a 2D texture we can go to very high
>> resolutions.
>>
>> 3) Pyramidal displacement mapping also tries to give the same quality as
>> relief mapping, but it doesnot require as large number of iterations as
>> relief mapping. here also we can go to very high resolutions as we are using
>> 2D textures.
>>
>> even the performance(fps) given by the 3 techiques is hugely different. i
>> will soon upload a comparison table for comparing their performances.
>>
>> ravish
>>
>>
>>
>> On Sat, Jul 12, 2008 at 6:26 AM, Darien Caldwell <
>> darien.caldwell at gmail.com> wrote:
>>
>>> Looking at the original Wiki link, I see there are examples of Relief
>>> Mapping, Sphere Tracing, and Pyramidal Mapping.  However looking
>>> closely, I can't see any visual differences.  What is the main
>>> difference between the 3 styles of mapping?
>>>
>>> On 7/11/08, Lawson English <lenglish5 at cox.net> wrote:
>>> > Karl Stiefvater wrote:
>>> >>> Again, I may be missing something - but this is how I originally
>>> >>> expected sculpts to work when they were initially described.
>>> >>
>>> >> heh... i think maybe you have too high of expectations.  :)
>>> >>
>>> >> this technology is still EXTREMELY young and experimental.  remember
>>> >> that when pixar puts this level of detail in a movie - each shot takes
>>> >> days of rendering time.  in a realtime engine - these things have
>>> >> never before been possible.
>>> >>
>>> >> but you do raise an interesting idea: perhaps the best way to quickly
>>> >> integrate mesostructure technology into our system is via the sculpt
>>> >> map - the artist provides a high resolution sculpt map - which still
>>> >> creates the standard low resolution mesh - but the details are
>>> >> fleshed-out via mesostructures.....
>>> >>
>>> >> very interesting...
>>> >
>>> > Of course, sculpt maps only have one collion model right now, anyway.
>>> > Torus? Sphere? Can't remember.
>>> >
>>> > Lawson
>>> > _______________________________________________
>>> > Policies and (un)subscribe information available here:
>>> > http://wiki.secondlife.com/wiki/SLDev
>>> > Please read the policies before posting to keep unmoderated posting
>>> > privileges
>>> >
>>>
>>
>>
>> _______________________________________________
>> Policies and (un)subscribe information available here:
>> http://wiki.secondlife.com/wiki/SLDev
>> Please read the policies before posting to keep unmoderated posting
>> privileges
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.secondlife.com/pipermail/sldev/attachments/20080713/31dbd8e6/attachment.htm