Rendering hair with graphics hardware

1
Rendering hair with graphics hardware

• Tae-Yong Kim
• Rhythm Hues Studios
• tae_at_rhythm.com

2
Overview

• State of the Art in Hair Rendering and Modeling
• Issues in Hardware Hair Rendering
• Line drawing with Graphics Hardware
• Illumination and Hardware Shader
• Shadows

3
Hair representation

• Hair is not a single entity always consider
  hair volume a a whole
• Surface (polygons, nurbs, etc.)
• Line (guide hair, curve, polyline..)
• Volumetric representation (implicit surface, 3D
  texture, tubes, )

4
Representation - Polygons

• Real-time rendering
• Suited for existing artists asset
• Limited hairstyle
• Animation difficult

ATI2004 - Polygonal Model (from SIGGRAPH Sketch
by Thorsten Scheuermann)
5
Representation - Lines

• Dynamic hair
• No restriction in hairstyle
• Advanced rendering possible
• No standard modeling technique
• Can be time consuming to render

nVidia Line representation (from Matthias
Wlokas Eurographics 2004 tutorial)
6
Model Generation

• Use control tools to generate bunch of hairs
• Control curves (guide hairs)
• Cylinders single level/multi level
• Fluid flow, 3D vector field
• Automated methods emerging

7
Guide hair
8
Other guide objects (fluid)
Hadap and Nadia M. Thalmann Eurographics 2001
9
Other guide objects (cylinders)
Kim and Neumann SIGGRAPH 2002
10
Emerging techniques
S. Paris, H. Briceno, F. Sillion SIGGRAPH 2004
11
Hair as line drawing
12
Hair segment as a GL line
DrawHairLine(ps, pe , cs, ce) glBegin(GL_LINE
S) glColor3fv(cs) glVertex3fv(ps) glColor3
fv(ce) glVertex3fv(pe) glEnd()
13
Hair as GL lines
DrawHair(p0,p1,..,pn-1,c0,c1,.cn-1) glBegin(
GL_LINE_STRIP) glColor3fv(c0) glVertex3fv(p0)
glColor3fv(c1) glVertex3fv(p1)
glColor3fv(cn-1) glVertex(pn-1) glEnd()
14
Are we done?
DrawHair(p0,p1,..,pn-1,c0,c1,.cn-1) glBegin(
GL_LINE_STRIP) glColor3fv(c0) glVertex3fv(p0)
glColor3fv(c1) glVertex3fv(p1)
glColor3fv(cn-1) glVertex(pn-1) glEnd()

Algorithm1 For each hair, call this function!
15
Not quite..
16
Point samples
True sample
Computed sample color
17
(No Transcript)
18

• Increase sampling rate ( large image size,
  accumulation buffer)
• Image quality depends on number of samples at a
  slow convergence rate
• Required sampling rate is above 10,000 x 10,000
  pixel resolution
• Thinner (smaller) hairs require even higher
  sampling rate

19

• Hardware antialiasing of line drawing
• GL_LINE_SMOOTH
• Thickness control with alpha blending

20
? 0.09 0.25 0.60
1.0
21
Correct
22
Hair as visibility ordered lines
Algorithm 2 1. Compute color for each end point
of the lines (shading, shadowing) 2. Compute the
visibility order 3. Draw lines sorted by the
visibility order
23
a
o
b
r
k
v
c
e
j
p
g
d,e
s
m
d
c
i
f,g,h
u
l
q
h
n
b
a
f
i,j
t
k,l,m
n,o,p,q
r,s,t
u,v
Drawing order a, b, c, d, e, f, g, h, I, j, k,
l, m, n, o, p, q, r, s, t, u, v
24
(No Transcript)
25

• Visibility ordering can be cached and reused
  (useful for interactive application)

26

• Describes the amount of reflected/scattered light
  toward the viewing direction
• Kajiya-Kay (1989)
• Marschner et al. (2003)

27
V
L
T
28
H
V
L
T
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
Line table
Vertex table
35
Self-shadows

• A shading model describes the amount of
  reflected light when hair is fully lit
• Most hair receives attenuated light due to
  self-shadowing
• Crucial for depicting volumetric structure for
  hair

36
(No Transcript)
37
(No Transcript)
38

• Shadow is a fractional visibility function
• ?How many hairs between me and the light?
• ?What percentage of light is blocked?

39

• Shadow is a fractional visibility function

40

• Opacity shadow maps
• Kim and Neumann EGRW 2001
• Fast approximation of the deep shadowing function
• Idea use graphics hardware as much as possible

41
Opacity
O(l)
Monotonically increasing
l
42
(No Transcript)
43
for (1 ? i ? N) Determine the opacity
maps depth Di from the light for each shadow
sample point pj in P (1 ? j ? M) Find i
such that Di-1? Depth(pj) lt Di Add the
point pj to Pi. Clear the alpha
buffer and the opacity maps Bprev, Bcurrent.
for (1 ? i ? N) Swap Bprev and
Bcurrent. Render the scene clipping
it with Di-1 and Di. Read back the alpha
buffer to Bcurrent. for each shadow sample
point pk in Pi O prev sample(Bprev ,
pk) O current sample(Bcurrent ,
pk) O interpolate (Depth(pk),
Di-1, Di, O prev, O current) t(pk)
e-?O F(pk) 1.0 - t(pk)
44
Uniform slicing
1D BSP
Nonlinear spacing
45
for (1 ? i ? N) Determine the opacity
maps depth Di from the light for each shadow
sample point pj in P (1 ? j ? M) Find i
such that Di-1? Depth(pj) lt Di Add the
point pj to Pi. Clear the alpha
buffer and the opacity maps Bprev, Bcurrent.
for (1 ? i ? N) Swap Bprev and
Bcurrent. Render the scene clipping
it with Di-1 and Di. Read back the alpha
buffer to Bcurrent. for each shadow sample
point pk in Pi O prev sample(Bprev ,
pk) O current sample(Bcurrent ,
pk) O interpolate (Depth(pk),
Di-1, Di, O prev, O current) t(pk)
e-?O F(pk) 1.0 - t(pk)
46

• Storing all the opacity maps incur high memory
  usage
• Sort shadow computation points based on the maps
  depth

P0
Pi
PN-1

• As soon as the current map is rendered, compute
  shadows for corresponding sample points.

47
for (1 ? i ? N) Determine the opacity
maps depth Di from the light for each shadow
sample point pj in P (1 ? j ? M) Find i
such that Di-1? Depth(pj) lt Di Add the
point pj to Pi. Clear the alpha
buffer and the opacity maps Bprev, Bcurrent.
for (1 ? i ? N) Swap Bprev and
Bcurrent. Render the scene clipping
it with Di-1 and Di. Read back the alpha
buffer to Bcurrent. for each shadow sample
point pk in Pi O prev sample(Bprev ,
pk) O current sample(Bcurrent ,
pk) O interpolate (Depth(pk),
Di-1, Di, O prev, O current) t(pk)
e-?O F(pk) 1.0 - t(pk)
48

• The alpha buffer is accumulated each time the
  scene is drawn.
• The scene is clipped with Di and Di-1
• Speedup factor of 1.5 to 2.0
• In very complex scenes, preorder the scene
  geometry so that the scene object is rendered
  only for a small number of maps.
• ?More speedup and reduce memory requirement for
  scene graph

49
for (1 ? i ? N) Determine the opacity
maps depth Di from the light for each shadow
sample point pj in P (1 ? j ? M) Find i
such that Di-1? Depth(pj) lt Di Add the
point pj to Pi. Clear the alpha
buffer and the opacity maps Bprev, Bcurrent.
for (1 ? i ? N) Swap Bprev and
Bcurrent. Render the scene clipping
it with Di-1 and Di. Read back the alpha
buffer to Bcurrent. for each shadow sample
point pk in Pi O prev sample(Bprev ,
pk) O current sample(Bcurrent ,
pk) O interpolate (Depth(pk),
Di-1, Di, O prev, O current) t(pk)
e-?O F(pk) 1.0 - t(pk)
50
(No Transcript)
51
for (1 ? i ? N) Determine the opacity
maps depth Di from the light for each shadow
sample point pj in P (1 ? j ? M) Find i
such that Di-1? Depth(pj) lt Di Add the
point pj to Pi. Clear the alpha
buffer and the opacity maps Bprev, Bcurrent.
for (1 ? i ? N) Swap Bprev and
Bcurrent. Render the scene clipping
it with Di-1 and Di. Read back the alpha
buffer to Bcurrent. for each shadow sample
point pk in Pi O prev sample(Bprev ,
pk) O current sample(Bcurrent ,
pk) O interpolate (Depth(pk),
Di-1, Di, O prev, O current) t(pk)
e-?O F(pk) 1.0 - t(pk)
52

• Quantization in alpha buffer limits O to be 1.0
  at maximum
• ? scales the exponential function s. t. O value
  of 1.0 represents a complete opaqueness (t 0)
• ? 5.56 for 8 bit alpha buffer
• (e-? 2-8)

1.0
1.0
1.0
53
(No Transcript)
54

• Setup pass
• Compute the visibility order
• Compute shadow values
• Drawing pass
• For each line segment Li ordered due to the
  visibility order
• Set thickness (alpha value)
• Draw Li with programmable shader

55

• Whole opacity maps stored in 3D texture
• nVidias demo
• Koster et al., Real-Time Rendering of Human Hair
  using Programmable Graphics Hardware, CGI 2004
• Mertens et al, A Self-Shadow Algorithm for
  Dynamic Hair using Density Clustering, EGSR 2004

56

• Antialiasing
• Shading through vertex/fragment shader
• Shadows with opacity maps

57
(No Transcript)