Plane by Python Generator

Cairyn

Just food for thought:
I once had the same desire to create landscapes in optimum resolution, but I came to realize that it is necessary to create a dynamic subdivision that reserves the higher resolutions for spots close to the camera. The numbers just accumulate too fast:

Subdivision of 15 means
2^15 = 32768 segments in either direction, which amounts to
32768^2 = 1,073,741,824 polygons (one gigapoly ).
Almost the same number of points is needed to define these polygons
(strictly one row and one line more, so 32769^2 = 1,073,807,361).
Each point is a vector from three floats, which is eight bytes each, resulting in a payload of roundabout 8GB for storing the points alone. Each polygon refers to four points which requires an index of four bytes at least (too lazy to look up the type now), which is another 16GB.

Thus, your plane has a memory footprint of 24GB for (2^15)^2 segments (for 2^17 it's 16 times more, or 384GB). A well equipped machine may still be able to work with that, but it's not practical, and causes inavoidable lags and most likely swapping. Not having a special datatype underlying, C4D may need to copy that data for rendering, too...

indexofrefraction

@ Cairyn

i know that this hogs down c4d at some point.
that is exactly why i want a dynamically adjustable object, not a fixed subdivided polygon object
like this i can test the size/subdivision my machine can handle

of course it would be even better to have a plane with a polygon-resolution-fall off
but one step at the time

for now it would be interesting why c4d lags that fast...
i suppose there is something not good in this code,...

Manuel

Hello,

mmm Ocean

as @Cairyn mention, this way is really too "brute force"
If on top of that you want to add a deformer to simulate the ocean, you are going to kill your computer.

You should search how you can build a mesh taking into account the screen/camera space. There's no point into having that kind of details km away from the camera. (I'm sure you are already aware of that)

You should try to build your own geometry instead of using the modeling command subdivide.

other ideas :

you don't want to go with a plugins because of the render farm but remember that you can still export the animated mesh with alembic. Of course that will create a huge file, but will reduce the render time.

You can also have a "light" grid using one of the Ocean deformer that are available out there and use a displacement map (you can have some feedback of that on the viewport with the viewport tessellation).

This is really shot depending also.

Cheers
Manuel

indexofrefraction

hm.. ok...

i guess what would be best is an adaptive plane....

outer size, ie 5km
center size, ie 500m
outer segment size, 10m
center segment size, 10cm
falloff type (linear/exponential/...)

i still wonder if the code above is good,
or if it hogs down because of event loops

Manuel

hello,

The generator is already caching your object. That what "optimize cache" checkbox is for, you don't have to deal with CheckDirty and things like that.
If the generator see nothing different, it will send the cache. If you change a parameter, it will rebuild the object (execute the code, update the cache etc)

Try to pick a simple plane and launch the subdivide command by hand, you will see that after some, the command will start to be slow. Go step by step, start with one single polygon, and subdivide by 1, after 16 777 216 polygons, you will see the command take some times to go to 67 108 864 poly.

Cinema 4D should start to show that he's not happy. And it's not a generator.

Your code doesn't loop or anything, you are just asking for too much.

Cheers
Manuel

FlavioDiniz

@indexofrefraction I know this is not what you've asked, but it might help you:
Redshift does have a method of subdivision called screen space adaptive, which make more subdivisions on polygons that are close to the camera and less to those that are far from the camera. So it relatively lightweight because it happen only at render time and still bring good results.
alt text

indexofrefraction

thank you guys,

im intrigued to make something like this as a Python Generator

starting in the center
generate a square of x square polygons
then at some point double the polygon size
generate x borders with the new polygon size
etc etc

not such an easy algorithm, tough

... isnt there a built in solution for something like that?
adaptive / falloff subdivision ....

Manuel

hello,

There's no such a function as we can tell in the sdk.

There's not one single solution for your problem. As it's for Ocean Simulation your problem is more 2D than 3D. You need to build a plan.

You can create points in your space and use a 2D Delaunay triangulation, that could work.
You can create a 2D grid on screen space and project that grid into world space.

I don't know witch one works and witch one could be the fastest.

a very naive way to go is this one but maybe that will help you to try to find formula and start somewhere.

import c4d
#Welcome to the world of Python


def main():
    # Width
    l = 5000
    # Height
    L = 5000

    # Segments per lengh
    lSegments = 20
    LSegments = 10

    # Space beetween points
    lSpace = l /(lSegments - 1)
    LSpace = L /(LSegments - 1)

    # halfPoint
    lhalfSegment = lSegments * 0.5

    # Creates points
    points = []
    for z in xrange(LSegments):
        for x in xrange (lSegments):
            p = c4d.Vector(0.0)
            # This is where you can change the formula to change point position
            p.x = lSpace * ( x - lhalfSegment) 
            p.z = LSpace * z
            points.append(p)

    # Creates the polygon
    pol = c4d.BaseObject(c4d.Opolygon)

    # Points number and polyCount
    ptCnt = lSegments * LSegments
    polyCnt =  (lSegments - 1 )* (LSegments - 1)

    # Resize Object
    pol.ResizeObject(ptCnt, polyCnt)

    # Updates points position
    pol.SetAllPoints(points)
    polyIndex = 0
    
    # Creates CPolygon structure for each polygon (points ID for each polygon)
    for x in xrange(lSegments - 1):
        for y in xrange (LSegments - 1):
            xmove = x + y * lSegments
            ymove =  x + y * lSegments + lSegments
            cpol = c4d.CPolygon( xmove ,ymove ,ymove + 1 ,xmove + 1 )    
            pol.SetPolygon(polyIndex , cpol)
            polyIndex += 1
    # Updates the polygon object
    pol.Message(c4d.MSG_UPDATE)    

    # Done
    return pol

Cheers
Manuel

indexofrefraction

thank you manuel,

atm i'm going another way...

make a plane with a low subdivision
select polygons with a distance from the center
subdivide
repeat at 2. with lower distance
etc...

i guess this is probably faster anyway
(compared to setting up all points manually)

ps.
adjacent polygons are not sharing the same points in your code, right?
this is the tricky thing with setting up a plane, i guess .)

Manuel

hello,

yea you can try this way also, but selecting polygons closer to camera will take some time because there will be more and more polygons. But that can do the trick.

@indexofrefraction said in Plane by Python Generator:

adjacent polygons are not sharing the same points in your code, right?
this is the tricky thing with setting up a plane, i guess .)

Of course they are. But there's no UVs.

Cheers
Manuel

indexofrefraction

still working on it, it looks very promising...

generating the geometry seems to work fast enough, but...

adding a phong tag to the python generator has no effect
the solution for this is here:
https://developers.maxon.net/forum/topic/10383/13881_python-generator--phong-tag/2
but would it be possible to get UVs for such a plane?
c4d.utils.GenerateUVW() ? TempUVHandle ?

EDIT:
similar to the phong tag solution i tried to add an uvw tag to the generators cache object with:

    uvw = c4d.BaseTag(c4d.Tuvw)
    polygon.InsertTag(uvw)

now, if i convert the generator to polygons I get:

A problem with this project has been detected:
Object "OceanPlane" - Tag 5671 not in sync.
Please save and contact MAXON Support
with a description of the last used commands, actions or plugins.

the object then has an uv tag, tough,
but mapping does not work, yet.

any tips to get working uv coordinates on such a plane object?

Manuel

hello,

I've updated my code, there were some bugs with type and things like that.
Also added the phong tag and a way to create a UVW tag in the most easiest way i found.

import c4d
#Welcome to the world of Python


def main():
    # Width
    l = 500
    # Height
    L = 500

    # Segments per lengh
    lSegments = 500
    LSegments = 500

    # Space beetween points
    lSpace = l /(lSegments - 1.0)
    LSpace = L /(LSegments - 1.0)

    # halfPoint
    lhalf = l * 0.5

    # Creates points
    points = []
    for z in xrange(LSegments):
        for x in xrange (lSegments):
            p = c4d.Vector(0.0)
            # This is where you can change the formula to change point position
            p.x = lSpace * x - lhalf
            p.z = LSpace * z
            points.append(p)

    # Creates the polygon
    pol = c4d.BaseObject(c4d.Opolygon)

    # Points number and polyCount
    ptCnt = lSegments * LSegments
    polyCnt =  (lSegments - 1 )* (LSegments - 1)

    # Resize Object
    pol.ResizeObject(ptCnt, polyCnt)

    # Updates points position
    pol.SetAllPoints(points)
    polyIndex = 0

    # Creates CPolygon structure for each polygon (points ID for each polygon)
    for x in xrange(lSegments - 1):
        for y in xrange (LSegments - 1):
            xmove = x + y * lSegments
            ymove =  x + y * lSegments + lSegments
            cpol = c4d.CPolygon( xmove ,ymove ,ymove + 1 ,xmove + 1 )
            pol.SetPolygon(polyIndex , cpol)
            polyIndex += 1

    # Adds a phong tag (copy from generator or create a new one)
    # Checks for phong tag on python generator
    phong = op.GetTag(c4d.Tphong)
    if phong is None:
        # Creates one, if non-existent (careful with such operations, make sure, the tag is only created once)
        phong = op.MakeTag(c4d.Tphong)
    if phong is not None:
        pol.InsertTag(phong.GetClone()) # important to insert a clone of the original phong tag



    # Creates a new texture tag
    matTag = c4d.BaseTag(c4d.Ttexture)
    pol.InsertTag(matTag)

    # Changes the settings of that texture tag to cubic
    matTag[c4d.TEXTURETAG_PROJECTION] = c4d.TEXTURETAG_PROJECTION_CUBIC
    matTag[c4d.TEXTURETAG_POSITION] =c4d.Vector(0 , 0, L * 0.5  )
    matTag[c4d.TEXTURETAG_SIZE,c4d.VECTOR_X] = l * 0.5
    matTag[c4d.TEXTURETAG_SIZE,c4d.VECTOR_Z] = L * 0.5

    # Generates uvwCoordinates from the texture tag
    uvwTag = c4d.utils.GenerateUVW(pol, pol.GetMg(), matTag, pol.GetMg())
    # Inserts the uvwtag
    pol.InsertTag(uvwTag)

    # Removes the TEXTURE tag so we can add one on the generator.
    matTag.Remove()

    # Updates the polygon object
    pol.Message(c4d.MSG_UPDATE)

    return pol

Cheers
Manuel

indexofrefraction

Hey thanks ALOT, Manuel !

i'm creating the plane in a different way,
because i double the polygon sizes with distance
still, i must check your code for learning!

the phong & material tag solution is gold

EDIT:
all works like a charm, now
topic solved!

Manuel

hiya,

Feel free to share your code (if you want of course) as it could help other people.

If you think your question is solved, please change the states of this thread to solved.

Cheers
Manuel

mp5gosu

Yes, I'd like to see it of course. Maybe you can share it via Github so others can for and contribute to it.