Samiam’s Scribble Pad

October 31, 2010

Python Implementation of Spherical Harmonics Stratified Sampling

Filed under: Uncategorized — admin @ 7:32 pm

Its Sunday afternoon and its time to write some code, this is pretty much lifted verbatim from Robin Green’s 2003 Paper: Spherical Harmonic Lighting the Gritty Details

So for the next trick is to put this as a large table into a shader so I can bake out a set of 9 cooefficients (4 bands) for Shadowed Diffuse Transfer, these I will store per sample in a point cloud to be looked up during a shading stage.

Then after that I can implement the Image Based Lights, then I can do really quick Image Based lighting using Spherical Harmonics

Lots of fun!

I made a big mess of this but luckily Markus Kransler was able to fix it up:

Here is the amended code:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85

#!/usr/bin/env python 
 
class SHSample():
	sph=(0.0,0.0)
	vec=(0.0,0.0,0.0)
	coeff={}
	pass
 
def P(l,m,x):
	import math
	#Associated Legendre Polynomial P(l,m,x) at x
	pmm = 1.0
	if m > 0:
		somx2=math.sqrt(1.0-(x*x))
		fact = 1.0
		for i in xrange(1,m+1,1):
			pmm *= (-fact)*somx2
			fact += 2.0
	if l == m:
		return pmm
 
	pmmp1 = x * ((2.0*m)+1.0)*pmm
 
	if l == m+1:
		return pmmp1
 
	plm = 0.0
 
	for ll in xrange(m+2,l+1,1):
		plm = ((2.0*ll - 1.0) * x * pmmp1 - (ll + m - 1.0) * pmm) / (ll - m);
		pmm = pmmp1
		pmmp1 = plm
 
	return plm
 
def K(l,m):
	import math
	temp = float((((2.0*l)+1.0)*math.factorial(l-m))/(4.0*math.pi*math.factorial(l+m)))
	return math.sqrt(temp)
 
def SH(l,m,theta,phi):
	import math
	sqrt2 = math.sqrt(2.0)
	if m==0:
		return K(l,0)*P(l,0,math.cos(theta))
	elif m > 0:
		return sqrt2*K(l,m)*math.cos(m*phi)*P(l,m,math.cos(theta))
	else:
		return sqrt2*K(l,-m)*math.sin(-m*phi)*P(l,-m,math.cos(theta))
 
 
def setupSamples(sqrtNumSamples=64,numBands=4):
	import random,math
	counter = 0
	oneOverN = 1.0/float(sqrtNumSamples)
	samples = [SHSample() for i in range(sqrtNumSamples*sqrtNumSamples)]
 
	for i in range(sqrtNumSamples):
		for j in range(sqrtNumSamples):
			x = (i+ random.random())*oneOverN
			theta = 2.0*math.acos(math.sqrt(1-x))
 
			y = (j+ random.random())*oneOverN
			phi = 2.0*math.pi*y
 
			samples[counter].sph=(theta,phi)
 
			vec = (math.sin(theta)*math.cos(phi),\
			math.sin(theta)*math.sin(phi),\
			math.cos(theta))
			samples[counter].vec = vec
 
			tmpDict = {}
			for l in range(numBands):
				for m in xrange(-l,l+1,1):
					index = l*(l+1)+m
					sh= SH(l,m,theta,phi)
					tmpDict[index]=sh
			samples[counter].coeff=tmpDict
			counter +=1
	return samples
 
 
for i in setupSamples():
	print i.coeff

October 19, 2010

Some papers to read and implement

Filed under: Uncategorized — admin @ 12:51 pm

October 14, 2010

Point Cloud in Nuke using “PositionToPoints”, 3delight Rendman Shader Language and cortex-vfx

Filed under: Uncategorized — Tags: , , , , — admin @ 1:51 pm

Here is the end result:

Nuke Node "PositionToPoints" with 3d EXR inputs

First things first I need a model to work with

Model

Shader

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

surface bakeColourAndPosition(
uniform float diffuseAmount = 1;
varying color surfaceColour = color(0.18,0.18,0.18);
varying color opacityColour = color(0.99,0.99,0.99);
uniform string bakeFile="/tmp/out.bake";
)
{
   varying normal Nn = normalize(N);
   Ci = diffuse(Nn)*surfaceColour*diffuseAmount*Cs;
   Oi = opacityColour*Os;
   varying point Pworld = transform("current","world",P);
   bake(concat(bakeFile,"Position"),s,t,Pworld);
   bake(concat(bakeFile,"Colour"),s,t,Ci);
   Ci *= Oi;
}

This shader will produce two text “bakefile” files in the /tmp directory

Note: The texture coordinates are ignored, only using the 3rd,4th and 5th values

  1. one named out.bakeColour with colour information
  2. one named out.bakePosition with position information

Output Bakefiles

  • Download .tar.gz here…

    Due to the SIMD nature of shaders the line count of each of the bake files is the same so they should contain line for line the same information about the Position and Colour Respectively.

    They are in ASCII format so they are easy enough to parse with Python.

    here is an example of thier content:

    1
2
3
4
5
6
7
8
9

    out.bakePositionmh
3
0 1 -0.07415867 0.17987273 -0.05079475
0 1 -0.073529155 0.1800126 -0.051191031
0 1 -0.07289961 0.18015243 -0.051587344
0 1 -0.072270096 0.18029229 -0.051983685
0 1 -0.07164058 0.18043211 -0.052379965
0 1 -0.07101102 0.18057197 -0.052776248
0 1 -0.07038155 0.1807118 -0.053172619

    Creating Position and Colour EXR files using cortex-vfx

    1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77

    #!/usr/bin/env python
 
import sys,os,math
 
IECoreInstallPath = "/usr/lib/python2.6/site-packages"
 
if IECoreInstallPath not in sys.path:
	sys.path.append(IECoreInstallPath)
 
from IECore import *
 
bakeFolder = "/tmp"
 
colorBakeFileLocation = os.path.sep.join([bakeFolder,"out.bakeColour"])
 
positionBakeFileLocation = os.path.sep.join([bakeFolder,"out.bakePosition"])
 
 
 
 
def parseBakeFile(bakeFileLocation):
	data = []
	counter = 0
	bakeFile = open(bakeFileLocation,"r")
	for line in bakeFile.readlines():
		counter +=1
		if counter > 2:
			stuff = line.strip().split(" ")
			if len(stuff) > 2:
				data.append((float(stuff[2]),float(stuff[3]),float(stuff[4])))
	print "Completed parsing %d lines of file %s" % (len(data),bakeFileLocation)
	bakeFile.close()
	return data
 
colourData = parseBakeFile(colorBakeFileLocation)
 
positionData = parseBakeFile(positionBakeFileLocation)
 
if len(colourData) == len(positionData):
	squareSize = int(math.sqrt(len(positionData))) +1
	print "Square Size: %d, Excess Pixels : %d" % (squareSize,squareSize*squareSize - len(colourData))
	width = squareSize
	height = squareSize
	x = FloatVectorData( width * height )
	y = FloatVectorData( width * height )
	z = FloatVectorData( width * height )
 
	r = FloatVectorData( width * height )
	g = FloatVectorData( width * height )
	b = FloatVectorData( width * height )
 
	for i in range(len(colourData)):
		r[i]=colourData[i][0]
		g[i]=colourData[i][1]
		b[i]=colourData[i][2]
		x[i]=positionData[i][0]
		y[i]=positionData[i][1]
		z[i]=positionData[i][2]
 
	boxColour = Box2i( V2i( 0, 0 ), V2i( width-1, height-1 ) )
	boxPosition = Box2i( V2i( 0, 0 ), V2i( width-1, height-1 ) )
 
	imageColour = ImagePrimitive( boxColour, boxColour )
	imagePosition = ImagePrimitive( boxPosition, boxPosition )
 
	imagePosition["R"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, x)
	imagePosition["G"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, y)
	imagePosition["B"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, z)
 
	imageColour["R"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, r)
	imageColour["G"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, g)
	imageColour["B"]= PrimitiveVariable( PrimitiveVariable.Interpolation.Vertex, b)
 
	writePosition = Writer.create( imagePosition, "/tmp/outPosition.exr" )
	writeColour = Writer.create( imageColour, "/tmp/outColour.exr" )
	writePosition.write()
	writeColour.write()

    See more about cortex-vfx on google code:

    Using Nuke to read the Position and Colour Data

    File > Script Command [X] > PositionToPoints

    File > Script Command > PositionToPoints

    If you werent able to create your own pair of EXRs you can download the pair here in .tar.bz2 format

    So then you just need to connect them up to the input nodes for the PositionToPoints 3d node as follows:

    Nuke Node "PositionToPoints" with 3d EXR inputs

    If you thought this was useful leave a comment, or if you thought it was stupid leave a comment about how to improve it

    Sam

Powered by WordPress