1. Digital Image Synthesis Yung-Yu Chuang 10/04/2007
Shapes
Digital Image Synthesis
Yung-Yu Chuang
10/04/2007
with slides by Pat Hanrahan

2. Announcements One break or two breaks.
Please do read the textbook and trace code.
LootAt issue.
No class on 10/18 because of attending CGW 2007.
Assignment #1 will be handed out today. Due three weeks later.

3. Shapes
One advantages of ray tracing is it can support various kinds of shapes as long as we can find ray-shape intersection.
Careful abstraction of geometric shapes is a key component for a ray tracer. Ideal candidate for object-oriented design. Scan conversion may not have such a neat interface.
All shape classes implement the same interface and the other parts of the ray tracer just use this interface without knowing the details about this shape.

4. Shapes Primitive=Shape+Material
Shape: raw geometry properties of the primitive, implements interface such as surface area and bounding box.
Source code in core/shape.* and shapes/*

5. Shapes pbrt provides the following shape plug-ins:
quadrics: sphere, cone, cylinder, disk, hyperboloid (雙曲面), paraboloid(拋物面) (surface described by quadratic polynomials in x, y, z)
triangle mesh
height field
NURBS
Loop subdivision surface
Some possible extensions: other subdivision schemes, fractals, CSG, point-sampled geometry

6. Shapes All shapes are defined in object coordinate space
class Shape : public ReferenceCounted {
public:
<Shape Interface> all are virtual functions
const Transform ObjectToWorld, WorldToObject;
const bool reverseOrientation,
transformSwapsHandedness; }

7. Shape interface: bounding
BBox ObjectBound() const=0;pure virtual function
BBox WorldBound() { left to individual shape
default implementation; can be overridden
return ObjectToWorld(ObjectBound()); }

8. Shape interface: intersecting
bool CanIntersect() returns whether this shape can do intersection test; if not, the shape must provide
void Refine(vector<Reference<Shape>>&refined)
examples include complex surfaces (which need to be tessellated first) or placeholders (which store geometry information in the disk)
bool Intersect(const Ray &ray,
float *tHit, DifferentialGeometry *dg)
bool IntersectP(const Ray &ray)
not pure virtual functions so that non-intersectable shapes don’t need to implement them; instead, a default implementation which prints error is provided.
in world space

9. Differential geometry
DifferentialGeometry: a self-contained representation for a particular point on a surface so that all the other operations in pbrt can be executed without referring to the original shape. It contains
Position
Parameterization (u,v)
Parametric derivatives
(dp/du, dp/dv)
Surface normal (derived from
(dp/du)x(dp/dv))
Derivatives of normals
Pointer to shape

10. Shape interface
float Area() useful when using as a area light
void GetShadingGeometry(
const Transform &obj2world,
const DifferentialGeometry &dg,
DifferentialGeometry *dgShading)
No back culling for that it doesn’t save much for ray tracing and it is not physically correct
for object instancing

11. Surfaces Implicit: F(x,y,z)=0 you can check
Explicit: (x(u,v),y(u,v),z(u,v))
you can enumerate
also called parametric
Quadrics

12. Sphere A sphere of radius r at the origin Implicit: x2+y2+z2-r2=0
Parametric: f(θ,ψ)
x=rsinθcosψ
y=rsinθsinψ
z=rcosθ
mapping f(u,v) over [0,1]2
ψ=uψmax
θ=θmin+v(θmax-θmin)
useful for texture mapping θ

13. Sphere

14. Sphere (construction)
class Sphere: public Shape { ……
private:
float radius;
float phiMax;
float zmin, zmax;
float thetaMin, thetaMax; }
Sphere(const Transform &o2w, bool ro,
float rad, float zmin, float zmax,
float phiMax);
Bounding box for sphere, only z clipping
thetas are derived from z

15. Intersection (algebraic solution)
Perform in object space, WorldToObject(r, &ray)
Assume that ray is normalized for a while
Step 1

16. Algebraic solution If (B2-4AC<0) then the ray misses the sphere
Step 2
If (B2-4AC<0) then the ray misses the sphere
Step 3
Calculate t0 and test if t0<0 (actually mint, maxt)
Step 4
Calculate t1 and test if t1<0
check the real source code in sphere.cpp

17. Quadric (in pbrt.h) inline bool Quadratic(float A, float B, float C,
float *t0, float *t1) {
// Find quadratic discriminant
float discrim = B * B - 4.f * A * C;
if (discrim < 0.) return false;
float rootDiscrim = sqrtf(discrim);
// Compute quadratic _t_ values
float q;
if (B < 0) q = -.5f * (B - rootDiscrim);
else q = -.5f * (B + rootDiscrim);
*t0 = q / A;
*t1 = C / q;
if (*t0 > *t1) swap(*t0, *t1);
return true; }

18. Why?
Cancellation error: devastating loss of precision when small numbers are computed from large numbers by addition or subtraction.
double x1 = ;
double x2 = ;
double y1 = ;
double y2 = ;
double z = (y1 - y2) / (x1 - x2); // 11.5
Catastrophic cancellation. Devastating loss of precision when small numbers are computed from large numbers
by addition or subtraction. For example if x and y agree in all but the last few digits, then if we compute z = x - y,
then z may only have a few digits of accuracy. If we subsequently use z in a calculation, then the result may only
have a few digits of accuracy.
As a first example, consider the following contrived code fragment
double x1 = ;
double x2 = ;
double y1 = ;
double y2 = ;
double z = (y1 - y2) / (x1 - x2);
System.out.println(z);
It is easy to compute the exact answer by hand as (4/10,000,000,000,000) / (4/100,000,000,000,000) = 10.
However, Java prints out result Even though all calculations are done using 15 digits of precision, the
result suddenly has only 1 digit of accuracy.

19. Range checking if (t0 > ray.maxt || t1 < ray.mint) return false;
float thit = t0;
if (t0 < ray.mint) {
thit = t1;
if (thit > ray.maxt) return false; }
...
phit = ray(thit);
phi = atan2f(phit.y, phit.x);
if (phi < 0.) phi += 2.f*M_PI;
// Test sphere intersection against clipping parameters
if (phit.z < zmin || phit.z > zmax || phi > phiMax) {
... // see if we should check another hit point

20. Geometric solution
1. Origin inside?

21. Geometric solution 2. find the closest point, t=-O‧D
if t<0 and O outside return false
D is normalized t t

22. Geometric solution 3. find the distance to the origin, d2=O2-t2
if s2=r2-d2<0 return false; t s d r r O

23. Geometric solution 4. calculate intersection distance,
if (origin outside) then t-s
else t+s t t s O s r d r d O

24. Sphere Have to test sphere intersection against clipping parameters
Fill in information for DifferentialGeometry
Position
Parameterization (u,v)
Parametric derivatives
Surface normal
Derivatives of normals
Pointer to shape

25. Partial sphere u=ψ/ψmax v=(θ-θmin)/ (θmax-θmin)
Partial derivatives (pp103 of textbook)
Area (pp107)

26. Cylinder

27. Cylinder

28. Cylinder (intersection)

29. Disk

30. Disk

31. Disk (intersection) h
h-Oz t Dz D

32. Other quadrics
paraboloid
hyperboloid
cone

33. Triangle mesh
The most commonly used shape. In pbrt, it can be supplied by users or tessellated from other shapes.
Some ray tracers only support triangle meshes.

34. Triangle mesh
class TriangleMesh : public Shape { …
int ntris, nverts;
int *vertexIndex;
Point *p;
Normal *n; per vertex
Vector *s; tangent
float *uvs; parameters }
vi[3*i]
vi[3*i+1]
vi[3*i+2] p
x,y,z …
Note that p is stored in world space to save transformations. n and s are in object space.

35. Triangle mesh
Pbrt calls Refine() when it encounters a shape that is not intersectable. (usually, refine is done in acceleration structure creation)
Void TriangleMesh::Refine(vector<Reference<Shape>>
&refined) {
for (int i = 0; i < ntris; ++i)
refined.push_back(new Triangle(ObjectToWorld,
reverseOrientation, (TriangleMesh *)this, i)); }
Triangle only stores a pointer to mesh and an index.

36. Ray triangle intersection
Intersect ray with plane
Check if point is inside triangle
In order to implement this algorithm we had to address these six issues.

37. Ray plane intersection
Algebraic Method
Substituting for P, we get:
Solution:
In order to implement this algorithm we had to address these six issues.

38. Ray triangle intersection I
Algebraic Method
For each side of triangle:
In order to implement this algorithm we had to address these six issues.
end

39. Ray triangle intersection II
Parametric Method
In order to implement this algorithm we had to address these six issues.

40. Ray triangle intersection III
In order to implement this algorithm we had to address these six issues.

41. Fast minimum storage intersection
a point on
the ray
a point inside
the triangle

42. Fast minimum storage intersectionD O V2 D V2 1 V1 V1 V0 1 V0

43. Fast minimum storage intersection

44. Fast minimum storage intersection
Cramer’s rule

45. Fast minimum storage intersection

46. Triangle P2
(u2, v2) P
(u,v) P0
(u0, v0) P1
(u1, v1)

47. Shading geometry if (!mesh->n && !mesh->s) { *dgShading = dg;
return; }
// compute Barycentric coordinate

48. Shading geometry Normal ns; Vector ss, ts; if (mesh->n)
ns = Normalize(obj2world(
b[0] * mesh->n[v[0]] +
b[1] * mesh->n[v[1]] +
b[2] * mesh->n[v[2]]));
else ns = dg.nn;
if (mesh->s)
ss = Normalize(obj2world(
b[0] * mesh->s[v[0]] +
b[1] * mesh->s[v[1]] +
b[2] * mesh->s[v[2]]));
else ss = Normalize(dg.dpdu);