COMP9018 - Advanced Graphics Welcome! Today: –Administrative stuff. –Introductory remarks. –Voting on course arrangements. –Basic graphics + OpenGL revision

COMP9018 - Advanced Graphics

• Welcome!• Today:

– Administrative stuff.– Introductory remarks. – Voting on course arrangements.– Basic graphics + OpenGL revision or

curves & surfaces.


Administrivia• Web page:

– http://www.cse.unsw.edu.au/~cs9018– Includes a notice board + message

board.– Tim & I will read/contribute regularly. – Also lots of links and stuff.


Administrivia• People

– Waleed (that's me) ([email protected])– Tim (the smart guy)

([email protected])– Who to e-mail:

• Enrolments: waleed• Software problems: waleed• Web issues: waleed• Lecture material: whoever presented it• Tricky questions: lambert• Lecturer-in-charge/complex admin: lambert


Attitude to the course• Graphics is HUGE!• No one person can grasp all of it• Tim & I know stuff but we don't know

it all: guides rather than experts


Fourth year• This is a fourth year

subject/graduate subject• So rules are a little different.

– Smaller numbers = • more flexibility• higher standard (we hope)• more interactivity


Things we’d like you to do• Common sense:

– Doing the best you can in the subject– Respecting fellow students– Shutting up when you're not supposed to talk– Participating– Doing the preparation/homework we ask you to– Not cheating (we'll come down HARD on cheats)– Actually showing up to appointments/not

expecting us to be available all the time


Some basic rules• Questions allowed at any time ... but

raise your hands and wait for us to ask

• Open to suggestions• Academic etiquette: You’re allowed

to borrow or copy, but when you do, ACKNOWLEDGE it.


Anybody scared?• Can discontinue without failure until

end of week 4 (I think)


Run through handout• See handouts


Voting issues• Vote on lots of critical things• ... but you can't vote people out of

the course• Course isn't going to get smaller and

smaller as time goes on, 'til last person gets an HD


Issue 1• Language preference

– C/C++ vs Java– Any other bizarre languages? Fortran,

Perl, Python, Haskell?

• Examples in both? • Assignments: both allowed


Issue 2• Last time had a local games

company give a seminar/demo• Couldn’t organise film person; but

will try again this year … if you are interested

• Vote?


Issue 3• Syllabus material• Rough outline of what was covered

last time• Any topics from the list people feel

strongly about?• What to drop to include?


Week 13 demo night• Put it in your calendars• No lectures week 13• From 5pm to 10pm• Can bring in own equipment if you like• Do you guys want me to ask outsiders

(e.g. industry people) to attend?


Text

• Graphics has 2 properties:– Diverse – Fast-moving

• Generally means textbooks get out of date quickly

• The Net is a very good resource. • See links on Web page for online info.


… but if you really want …• Textbooks

– Computer Graphics: Principles & Practice Brown et al. Known as “Old Testament”

– Advanced Animation & Rendering Techniques by Watt & Watt. Known as “New Testament”

– OpenGL Programming Guide by OpenGL ARB. Known as the “Red Book” or “Lego book”. Available online.


Labs• Have hardware acceleration (TNT2 cards

-- not great, but better than nothing)• Linux is official platform (not by choice,

really)• For compatibility: Only use standard Java

libraries + gl4java; for C/C++: standard libraries + OpenGL + GLU + GLUT + MUI ... must compile with gcc

• If you want something else, let us know.


Project possibilities• Write your own game• Write your own modeller• Extend an open source program with

some feature• Make a short CG movie• But be careful with your time

budget! Don’t want you to fail other subjects (almost happened last time)


Examples from last time• Some examples

– Movie scenes:• Lord of the Rings + modelling tools (4 ppl).• Story of a fly.

– Games:• 3D pool (2 ppl).• 3D tetris• 3D multiplayer networked robot/shooter/transfomer

game (2ppl).• Halflife level modelled on Ground & Basement of CSE. • Global thermonuclear war with Ronald Raygun,


More examples– Interactive applications

• Origami tutor• CT-Scan lung visualiser• Photo-etching animation.• Artificial creatures lab• Lego modeller

– Screen savers:• Rollercoaster generator.• Rubik’s snake.


Special resources• Generally

– Increased disk quota (+ 20Mb)– Increased IP quota (+ 200Mb)– Trying to arrange special access to

bongo lab.


Project-specific• With access to bongo:

– 6 Maya floating licenses. – VirTools. – May be able to arrange other tools, but

may be difficult.


Special resources• Available on request• Good for the project• Two clusters:

– vina: 18 machines, 1x866MHz, 256MB RAM, Linux

– tines: 17 machines, 2x450MHz, 512MB RAM, Linux


Software available• GLUT 3.7• OpenInventor• POVRay• Blue Moon Rendering Tools

(RenderMan compatible, radiosity support)

• Java3D• Blender


Break time• After break, revision of basic

graphics + OpenGLor

• Get straight into curves and surfaces.

• Vote!!


Graphics Revision• What is graphics?• About anything computer generated

that you see:– In print– On a monitor– On television– At the movies– At art exhibitions


Usual representation• Most common representation of an

image is a bitmap: A 2D array of pixels

• Each pixel holds a description of the colour of that point

• The bitmap that holds the current image on display is called the frame buffer


Representing colour• Usual way is as a combination of red green

and blue; but monochrome is also possible• Different approaches:

– Fixed set of colours (e.g. CGA, EGA). Colour represented by a bit value

– Each pixel stores an index into an array of customisable colours (e.g. VGA) aka colour index mode

– Each pixel directly represents its colour directly (e.g. HighColor, TrueColor) aka RGB mode.


Objects• Primitive objects in graphics:

– Points: A single coordinate.– Lines: usually a parametric

representation (a + tb; 0 < t < 1) or a pair of coordinates.

– Polygons: A sequence of points. Usually talk of closed polygons (end is same as beginning)


Concave polygon Convex polygon

But first ... what is a polygon?

• Sequence of points enclosing a plane.

• In 3D, we use polygons that only have one face ... there is no "back" of a polygon.

• How do you decide which way a polygon is facing?


A

B

C

D

Into the page

A

B

C

D

Out of the page

Using the normal• If not specified, use the right hand

rule.


Types of polygon• Nasty polygons:

– Non-planar: Not all vertices are in the same plane.

– Self-intersecting: Polygons that intersect themselves are problematic.

– Concave: Have internal angles greater than 180 degrees. Complicate lots of things.

• We'll assume we don't have any of these.


Transformations• Frequently want to represent "doing" things to

objects. Doing things = "transformations".• Many useful transformations can be

represented as matrices applied to vectors• Homogeneous matrices are used so that

translations can be represented as matrices.• Homogenous means 1 more dimension than

underlying space: 3x3 for 2D, 4x4 for 3D


Why homogenous?• Allows us to represent translations

as a matrix multiply -- making almost all interesting modelling transforms


Example transforms• Rotate about origin in 2D • Scale• Translate• Rotation about y axis in 3D?• Perspective • etc etc etc


Window to Viewport mapping

• Maps part of the 2D plane of interest onto the monitor

• Usually includes an inversion because monitors are upside down

• Except with OpenGL, which maintains a Cartesian representation


Clipping• Finding bits outside the screen• Why?

– Speed ... don't render what we don't need to render

– Avoid overwriting random bits of memory


Graphics pipelines - 2D• Start with description of 2D objects

(lines, points, polygon)• Apply transformations to position

them• Apply window-to-viewport mapping • Clip• Rasterise


OpenGL• Provides a general platform

independent graphics API that's a rich but thin and efficient layer over the hardware.

• About 150 commands• For real-time 2D and 3D point, line

and polygon based rendering• Also does 3D lighting and texturing


What OpenGL doesn't do• Doesn't deal with windows or I/O• Doesn't handle curves or curved

surfaces• Doesn't have a file format or handle

complicated objects - only points lines and polygons

• Doesn't create image files


Companion libraries to OpenGL

• GLU (GL utility): Does tesselation of curved surfaces, some more flexible projections, etc

• GLUT (GL utility toolkit): A platform-independent window and I/O system (also now includes MUI - micro user interface for menus, dials etc)

• WGL, GLX: Platform-specific window and I/O systems

• OpenInventor: Provides high-level scene graph libraries


OpenGL and Java• Too slow to have an OpenGL

implementation in Java.• So usual technique: use a Java native

interface to OpenGL and provide a Java interface to it.

• We’ll be using GL4Java


Why not use Java3D?• Java3D is a high-level scene graph

library • More like OpenInventor than OpenGL• Java3D is implemented on top of

OpenGL!• Have a look at Java3D later• Grrr ...


More about OpenGL• Funky thing about it: It's basically a

state machine - i.e. another computer• Most commands don't produce visible

output - they set up the state for what happens between glBegin and glEnd commands

• Interesting stuff happens between the glBegin and the glEnd


Open GL + C/Java• Anatomy of a simple OpenGL

program in C/GLUT - hello.c• ... from the OpenGL Programmer's

Guide• Java’s pretty much the same.


The matrix stack• How OpenGL does transformations• A perfect fit for scene graph

traversal• How matrix stacks work


Basic operations• glMatrixMode(Mode)sets the matrix

stack we want to use. For now, only concerned with GL_MODELVIEW.

• glPushMatrix()copies current top of matrix stack and pushes on top of stack.

• glPopMatrix()pops off the top of the stack

• glLoadIdentity()loads identity matrix onto top of stack


How matrix stack is used.• When you do glTranslatef(...)it

actually postmultiplies by the matrix on top of stack and puts it back on top of the stack.

• Means that last matrix op is done first.


Example - On blackboard• Initialise stack with

glMatrixMode(GL_MODELVIEW);glLoadIdentity();

• Scale by 2glScaled(1.0, 2.0, 1.0);

• Rotate around x by 90glRotated(90, 1, 0, 0);

• Push on stackglPushMatrix();


Example - cont'd• Translate by a

glTranslatef(a.x, a.y, a.z);• Draw a point on screen.

glBegin(GL_POINTS);glVertex3f(1,1,1);glEnd();would actually draw at S(2).R(90).Tr(a).(1,1,1)

• Pop off stackglPopMatrix();


The scene graph• A representation of the scene


World

Xwing1

R2D2 Pilot Torpedo

Left Arm Head

TIEFighter

Generalisation: A scene graph

• Can represent all objects in a scene as a graph Each node contains

geometry

Each edge storesa transformation


World

Xwing1 Xwing2

R2D2 Pilot Torpedo

Left Arm Head

TIEFighter

Scene graph vs Scene tree• Previous was a scene tree. Can

easily generalise to scene graph, e.g.


Why is this cool?• Because it allows us to have libraries

of objects.• It's hierarchical, so we can have

objects within objects.• When we change the transform one

node in the scene graph, all its children move, too! Good for animation.


Example 2: Arm• Arm-> Forearm -> Hand -> Fingers• Transforming arm also moves

forearm, hands, fingers.• Transforming forearm also

transforms hands & fingers.


Scene graphs and the matrix stack

• Can use matrix stack for traversing the scene graph

• Algorithm:– Start at root node. Render this node's

geometry– For each child:

• Push the stack• Apply the modelling transform• Render the node (including children)• Pop the stack


The scene graph• Why post multiply? • Because previous algorithm works

with it; and it represents the common case.


Viewing scene• Lots of pieces.• Projection (3D only)• Camera setting• Clipping• Visible surface determination (mostly

3D)• Window to viewport mapping


Projection• Two types:• Parallel: All objects are the same

size, regardless of distance• Perspective: Further away objects

are smaller• Both can be represented as matrices


Viewing• How set camera's position?• Lots of solid maths, but at end of the

day, you need:– A position for the camera (view

reference point)– A view-plane normal– An up vector

• Use this to construct the viewing matrix.


Clipping• 2D case is easy• 3D: Have to find the "view volume".

Area that can be seen. Also usually define a near and far plane.

• See demo viewvolpar.wrl and viewvolpersp.wrl


Visible surface determination

• Have to decide what goes in front of what.• Three common approaches in polygon

rendering:– Depth sorting– BSP Trees– Z Buffers

• Other approaches like raytracing handle it automatically.


The OpenGL approach• Viewing transform: Goes on bottom

of modelview stack. • Projection & clipping combined -

essentially, a direct description of the view volume in camera coordinate system.

• Visible surface determination: do-it-yourself, or Z buffer.


OpenGL Viewing Transform• gluLookat(eyex, eyey, eyez,

lookx, looky, lookz, upx, upy, upz)

• Eye position = VRP• VPN = eye -look• up = up• Remember: Goes at BOTTOM of

modelview stack• Think of it as an extra root node.


View volume definition• Depends on projection.• Several functions for defining:• glOrtho

– parallel projection is also known as orthographic projection

– glOrtho(left, right, bottom, top, near, far)• glFrustum

– The truncated pyramid is called a frustum– glFrustum(left, right, bottom, top, near, far)


Where do they go?• A special stack for the projection

matrix called GL_PROJECTION. • Usually only two deep. • So usual code is

glMatrixMode(GL_PROJECTION)glLoadIdentity()glOrtho(...)


Convenience functions• GLU library has some convenient variants:• gluOrtho2D(left, right, bottom, top) =

glOrtho(left, right, bottom, top, -1, 1).• Proof: See Mesa source code. • gluPerspective(fovy, aspect, near, far)• Maths behind it: OpenGL source code.


What do projections do?• The transform the view volume into

the canonical view volume.• Canonical = standard. • CVV is the standard cube.• Why? Easy to clip against; can be

done in hardware.


Window->Viewport• Window to viewport has its own matrix stack• glViewport(x, y, width, height)• WARNING: Viewport is not inverted as in java!• x, y are the lower left corners of the viewport

rectangle in pixels.• Note: in OpenGL, win->view does NOT kill the

z values; the z values are kept for the z buffer.


Z buffering• Set as part of the window

characteristics. • Handled by the windowing system;

e.g. GLUT or Java. • Called the depth buffer in OpenGL

terminology.• Must be cleared after every rendered

frame.


Lighting and shading• Different models.

– Ambient– Diffuse (uses Lambert's law)– Specular (includes diffuse component -

uses Phong illumination eqtn)


Different ways of rendering a scene

• Polygon rendering (like OpenGL)• Ray tracing (like POVRay)• Radiosity • Ray-trace/Radiosity hybrids (like

POVRay [experimental] and BMRT)• Polygon is the only realtime option

right now.


Polygon rendering ... cont'd

• Focus on polygon rendering for a while. • How to shade polygons?

– Flat shading: one normal per polygon, shade whole polygon one colour

– Gouraud shading: one normal per vertex (comes from model), interpolate colours between vertices

– Phong shading: one normal per vertex, interpolate normals between vertices


Light source and material• Lighting equations depend on:

– Light source intensities– Material properties


Light sources in OpenGL• glLightf(lightid,property,value)• All purpose lighting function• lightnum is the id of the light (GL_LIGHT0 to

GL_LIGHT7, but may be more). • Settable properties include:

– Light position (note! Modelview matrix applies!)– Diffuse intensity, ambient intensity, specular

intensity– Light attenuation with distance– Spotlight setting


Global properties• Global lighting is controlled by the

glLightModel() function• Properties include:

– Ambient light– Two-sided lighting– Local viewer


Material properties• glMaterial(face, property, value)• Applies to all subsequent values • face: front or back side of polygon• Properties:

– ambient, diffuse, specular and emissive colour components

– shininess (Phong exponent)


Other important functions• For polygons, must define normals.

So need glNormal() function• Just like material properties: normal

applies until you change it.


Texture mapping• Using images to add surface detail• Example: logopoly.wrl• Problem: How to "map" texture space to

polygon's surface.• Solution: Use texture coordinates for every

vertex and apply all that transformation stuff.

• More detail on texture mapping in Advanced OpenGL lecture.


Variants of texture mapping

• Procedural textures instead of images.• 3D vs 2D textures. • Using textures to model

– Surface perturbations: Bump map.– Reflections: Environment map.

• Textures are the "hot" area in graphics. Incredible number of nifty hacks possible using textures and variants.


Aliasing• Whenever we try to capture a high-

frequency (i.e. detail) reality with a low-frequency sample.

• Examples: Jaggies, texture collapse, etc.

• Approaches to fix: prefiltering, postfiltering, adaptive supersampling, jitter, mipmapping, bilinear interpolation, trilinear interpolation


Brains full yet?• Good!• Want something interesting to read?• Subject Web page: Readings - 19

Pitfalls to avoid when programming OpenGL; by Mark Kilgard: Person who wrote GLUT, and a couple of other OpenGL books, now works at nVidia.

Documents

COMP9018 - Advanced Graphics Welcome! Today: –Administrative stuff. –Introductory remarks. –Voting on course arrangements. –Basic graphics + OpenGL revision