Hi there,

after adding some automatic adaption of the level of detail my old PC was
able to render a complete fern. It looks quite confused since it is built
randomly.

Some details:

It consists of 22 wings with varying number of leaves. In total there were
8154 objects (there were more than 106000 in the full-detail version!).
Total computation time was 50 minutes and 57 seconds. Both, rendering and
parsing took almost exactly the half of this: 25min 36/21 sec. My PC is a
very old P200.

I am satisfied with the result therefore it is no longer of WIP status.
Thank you very much to all who gave me tips and hints!
Now I will clean-up the include files and release them as soon as possible.
It needs some user manual and a test-scene to be used properly.

Now I will concentrate on moss, waterdrops, snails and some stones. I think
some more pictures will follow here soon to get your comments.

live long and trace prosper, :-)
regards
Jürgen

Post a reply to this message

Attachments:
Download 'fern.jpg' (130 KB)

Preview of image 'fern.jpg'

"Jürgen Scharfy" <sch### [at] surfeuat> wrote :

>
> It consists of 22 wings with varying number of leaves.

    It might be possible to make a better plant, but you have it pretty good
right now. The individual sprays are great!

Post a reply to this message

Wow!

Pretty cool! :))

Congratulations. Looks Great!!!


Best regards.

Post a reply to this message

The looks very good! But perhaps it would look even better with some
different age variations in the leaves. The "wings" at the center for
example should be more curled. And at the very center you should have one or
two fiddle heads. Otherwise I love the detail!

Post a reply to this message

That looks very very groovy,

Congratulations!

Fernando.

Post a reply to this message

> The looks very good! But perhaps it would look even better with some
> different age variations in the leaves. The "wings" at the center for
> example should be more curled. And at the very center you should have one
or
> two fiddle heads. Otherwise I love the detail!

Your are right. As mentione I just stacked some wings together, randomly
arranged around a circle with some variation in the number of leaves and
their shape. Since I have to do some tests for the auto-detail-levelling I'm
shure there will be more pictures until I can release the include files.
If it is of interest for the group I can continue posting the results.
Unfortunately a test run takes about an hour...

Thanks for feedback!
regards
Jürgen

Post a reply to this message

I hope I can hold this level of realism when creating all the rest of the
scene....

Post a reply to this message

Pretty amazing.

--
#local X=20*<-2,2,5>;sphere_sweep{catmull_rom_spline 6<-8,-8>1<-8,-8>1<-8,8>1<
8,-8>1<8,8>1<8,8>1translate 20*z pigment{gradient z scale 3color_map{[0rgb<0,9
,18>][1rgb 0]}}}#local K=2*z*X-X;#local R=seed(clock);blob{#while(K.x>X.x)
#local N=X+<rand(R)rand(R)1>/3;#local X=(vlength(N-K)<vlength(X-K)?N:2*X-N);
sphere{X,1,1rotate z*90}sphere{X,1,1}#end pigment{rgbt 1}interior{media{//Nekar
emission<2,4,5>*5}}hollow scale.05}//   http://nekar_xenos.tripod.com/metanoia/
"Jürgen Scharfy" <sch### [at] surfeuat> wrote in message
news:3c4c6f05@news.povray.org...
> Hi there,
>
> after adding some automatic adaption of the level of detail my old PC was
> able to render a complete fern. It looks quite confused since it is built
> randomly.
>
> Some details:
>
> It consists of 22 wings with varying number of leaves. In total there were
> 8154 objects (there were more than 106000 in the full-detail version!).
> Total computation time was 50 minutes and 57 seconds. Both, rendering and
> parsing took almost exactly the half of this: 25min 36/21 sec. My PC is a
> very old P200.
>
> I am satisfied with the result therefore it is no longer of WIP status.
> Thank you very much to all who gave me tips and hints!
> Now I will clean-up the include files and release them as soon as possible.
> It needs some user manual and a test-scene to be used properly.
>
> Now I will concentrate on moss, waterdrops, snails and some stones. I think
> some more pictures will follow here soon to get your comments.
>
> live long and trace prosper, :-)
> regards
> Jürgen
>
>
>


---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.314 / Virus Database: 175 - Release Date: 2002/01/11

Post a reply to this message

"Jürgen Scharfy" wrote:
> 
> Hi there,
> 
> after adding some automatic adaption of the level of detail my old PC was
> able to render a complete fern. It looks quite confused since it is built
> randomly.
> 

It looks good, but some environment would much improve it, maybe place it
in some pot on a table or something similar.

> 
> It consists of 22 wings with varying number of leaves. In total there were
> 8154 objects (there were more than 106000 in the full-detail version!).
> Total computation time was 50 minutes and 57 seconds. Both, rendering and
> parsing took almost exactly the half of this: 25min 36/21 sec. My PC is a
> very old P200.
> 
> [...]

Again, if you make the whole fern one mesh (or at least each wing) this
would seriously speed up the render.  BTW as usual with such a geometry
generating function it would probably be useful to add file writing
routines that you can load the fern from file if you want to render it a
second time.  

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

> Again, if you make the whole fern one mesh (or at least each wing) this
> would seriously speed up the render.  BTW as usual with such a geometry

You already mentioned it and I followed already (partially) your suggestion.
Thanks for that.
Up to now each leaf is a single mesh (I merged the two meshed that built up
the leaf before).
Since the leaves are transformed after creation I would have to put all the
possible transformations into the mesh generation routine. A fern wing is a
really flexible macro that reacts on shape variables, age, "weakness" and
more. I don't want to put all these algorithms into the leaf macro since
they are only related to the wing (beside the size and age of the leaves).
That's not nice and less flexible. Additionally I like the object-oriented
apporach more than a putting-it-all-into-one-object mechanism. Of course on
the cost of rendering time.

But what is the difference? If I make single objects, POV builds a bounding
box around it at doesn't care about the object if the ray doesn't hit it.
This check is done on a pretty high level. Any sub-processing is skipped for
this object.
If it is a single mesh with a lot of holes in it (between the leaves) it
hits the bounding box all the time and has to analyse the mesh-triangles
before it decides to do nothing. Is that really faster?

> generating function it would probably be useful to add file writing
> routines that you can load the fern from file if you want to render it a
> second time.

That would be cool of course. I have first to get used to the file routine
stuff (if any) in POV 3.1g. AFAIK there is a lot of support for this in 3.5
beta but I don't want to download a new beta (which costs me almost an hour)
every 2 or 3 weeks.

Thanks a lot for you comments!
regards
SY

Post a reply to this message

On Wed, 23 Jan 2002 11:04:32 +0100, "Juergen Scharfy" <sch### [at] surfeuat>
wrote:
> That would be cool of course. I have first to get used to the file routine
> stuff (if any) in POV 3.1g. AFAIK there is a lot of support for this in 3.5
> beta but I don't want to download a new beta (which costs me almost an hour)
> every 2 or 3 weeks.

You can be interested in http://www.povray.org/ftp/pub/povray/MIRRORS.HTML

ABX

Post a reply to this message

> You can be interested in http://www.povray.org/ftp/pub/povray/MIRRORS.HTML

That's a list of mirror sites. The problem of the long download time is not
a slow server due to a wrong mirror. My problem is that I would have to
download 8MB through a 33kBit connection.
Or did I get you wrong?

best regards
SY

Post a reply to this message

On Wed, 23 Jan 2002 12:48:03 +0100, "Juergen Scharfy" <sch### [at] surfeuat>
wrote:
> > You can be interested in http://www.povray.org/ftp/pub/povray/MIRRORS.HTML
>
> That's a list of mirror sites. The problem of the long download time is not
> a slow server due to a wrong mirror. My problem is that I would have to
> download 8MB through a 33kBit connection.
> Or did I get you wrong?

No, you get me right. I didn't know reason of long download. If it is not your
case I hope at least it refreshed memory for others :-)

But I have another workaround for your long download. Install it once and later
download only second compiler version. It is little slower at work but only
1,1MB to download. You could download full package only when you find some bug
or read something important in
http://www.povray.org/binaries/windows/revision.txt 

ABX

Post a reply to this message

> But I have another workaround for your long download. Install it once and
later
> download only second compiler version. It is little slower at work but
only
> 1,1MB to download. You could download full package only when you find some
bug
> or read something important in

I see. Thanks for the tip!
Anyway, I just found out that I have all I need already in POV 3.1g so there
is no urgent need to download a new version.
I will implement file support although evaluating the functions isn't really
slow. The major advantage would be to share ferns via a definition file. The
slowest thing is creating the POV-objects since they allocate memory. This
has to be done anyway, no matter if the results are calculated or read from
file....

regards
SY

Post a reply to this message

Juergen Scharfy wrote:
> 
> [...]
> But what is the difference? If I make single objects, POV builds a bounding
> box around it at doesn't care about the object if the ray doesn't hit it.
> This check is done on a pretty high level. Any sub-processing is skipped for
> this object.
> If it is a single mesh with a lot of holes in it (between the leaves) it
> hits the bounding box all the time and has to analyse the mesh-triangles
> before it decides to do nothing. Is that really faster?

It is much faster unless you build a bounding box hierarchy yourself.  If
not, Povray always has to test against all the bounding boxes of several
thousand shapes which can be quite slow.  For a mesh it generates a very
efficient bounding structure (octree).  Apart from this the latter has
serious advantages if you want to render more than one fern

Transforming the vertices is not that much of a problem with
'vtransform()' but this is only available in megapov/Povray 3.5.

> That would be cool of course. I have first to get used to the file routine
> stuff (if any) in POV 3.1g. AFAIK there is a lot of support for this in 3.5
> beta but I don't want to download a new beta (which costs me almost an hour)
> every 2 or 3 weeks.

There is not much in Povray 3.5 to simplify this apart from some macros
like 'VStr()'.  You can find a good example how to do file writing in the
maketree include file.

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

Hi Christoph,

> It is much faster unless you build a bounding box hierarchy yourself.  If
> not, Povray always has to test against all the bounding boxes of several

I see. I will do some tests to get a comparison of these two methods. I
don't have a "feeling" how much the difference is in percent of total time.
This is very interesting. I let you know ASAP.

Thanks again,
regards
SY

Post a reply to this message

Hi there,

I just did some tests according to the comparison of a single, big mesh and
a structure of unions of small meshes.
The results are quite surprising.

I rendered a single wing with 40 leaves in 320x240 AA 0.3 in highest level
of detail (auto LOD settings turned off)
The picture was quite distorted since I removed any transformations beside a
single alterating horizontal shift of the leaves.

Single mesh variant:

85 objects (the stem was still included)
Parsing 1:20 (min:sec)
Tracing 1:04
Total 2:24

Complex unions:

11762 objects (full detail!)
Parsing 1:32 (only slight difference since code is unchanged there but some
more calculations)
Tracing 1:35 (slightly slower)
Total 3:07

As I could observe, in the union test (original code) tracing got faster to
the end. Obviously POV removed bounding boxes which coudn't be hit anymore
which  would subsequently reduce the number of boxes during tracing.
There were 138 times more objects (138837%)  in the second case but tracing
time increased only by 48%. 48% sounds a lot but the number of objects for a
complete fern (not only a wing) with auto LOD will have approx. 8000
objects!
I am shure that the single-mesh-solution has some advantages in case of a
complete fern since the bounding boxes are spread over the entire picture.

Anyway, the difference is (up to now) only 30 seconds. I decided to leave
the code as it is since I have to concentrate on the shape of the wings and
a macro assembling a whole fern. If parsing time gets more and more an issue
I can flip over to the single-mesh variant.
Thanks for your explanations,

best regards
SY

Post a reply to this message

"Jürgen Scharfy" wrote:
> 
> [...]
> 
> Single mesh variant:
> 
> 85 objects (the stem was still included)

Make the stem part of the mesh, i'm sure the stem objects (cylinders i
suppose) take more time than the mesh.  

And as you mentioned the advantages will increase with more complex
geometry.

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

> Make the stem part of the mesh, i'm sure the stem objects (cylinders i
> suppose) take more time than the mesh.

Cones to be exact since the stem gets thinner and thinner.
The stem itself (there was a part at the end where only stem cylinders were
visible and no more leaves) takes a second to trace. 99% of the time was
used for the leaves.
But I'm still working on POV 3.1g
regards
SY

Post a reply to this message

"Jürgen Scharfy" wrote:
> 
> Cones to be exact since the stem gets thinner and thinner.
> The stem itself (there was a part at the end where only stem cylinders were
> visible and no more leaves) takes a second to trace. 99% of the time was
> used for the leaves.

It seems there is something going wrong there, but without the seeing the
code i can't say anything definite.  

> But I'm still working on POV 3.1g

The mesh code has not changed that much since then.

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message