assignment 5 (ray casting) extensiongroups.csail.mit.edu/graphics/classes/6.837/f02/... ·...

Lecture 12: Thursday, 17 October 2002

Assignment 5 (Ray Casting) extensionEye ray generation; ray-primitive intersectionNew due date is Monday, 21 October at 5pm

Assignment 6 (Ray Tracing) extensionShadows, reflection, refraction, super-samplingNew due date is Monday, 28 October at 5pm

Adjusted office hours:Thursday (today):

Addy, 4-7pm, W20 Linux ClusterProf. Durand, 4-6pm, Bldg. 4 SGI Cluster

Friday:Prof. Teller, 3-5pm, W20 Linux Cluster

Monday:Addy, 10am-Noon, NE43-256Prof. Durand, 2-4pm, NE43-254

MIT 6.837 Computer Graphics Thursday, 17 October 2002 (L12)

Today

Planning for the second half of the termGetting started on your final team projects

Lecture: Ray TracingContrast scan-conversion, ray casting, ray tracingAppropriate physical units for ray tracingRay tracing pseudocode (Assignments 5 & 6)Ray tracing extensions


Final Projects

Start thinking about what you’d like to do!Animation? Game? Visualization? Algorithm? Interaction method? Etc.Representative final project reports linked from F99 course page

Team-building page, with list of past project titles, athttp://graphics.lcs.mit.edu/classes/6.837/F02/projects/teams.htm

Enter a brief description of your project idea or interests(along with your contact info, relevant background/skills, etc.) there

Next Thursday, in class:Brain-storming and team-building sessionCome with project ideas in mind; we’ll discuss as a groupStudents watching off-line: come to Prof.s’ office hours

Teams formed, project proposals due by November 1That’s two weeks from tomorrow !

Thereafter, weekly meetings with TA through end of term


Scan Conversion: Visibility Resolution

“Feed-forward” rendering method, one polygon at a timePolygons issued in “immediate mode” to graphics pipelineFull-frame, single raster, or single-pixel depth bufferVisibility (depth) information computed many times at each pixel

Limitations of scan-conversion method for visibility:Restricted to scan-convertable primitives (simple surface types)Requires polygonalization (& tessellation choice) for all primitivesCauses faceting (geometric) artifacts at silhouettes, interior


Scan Conversion: Shading Computation

Shading computed with a constant amount of stateTypically, small number of hardware-supported light sources

Local shading computation f (position, material, light, viewer)Evaluated at polygon vertices, Gouraud interpolated

Limitations of local shading computation & interpolation:Cartoonish look; incorrect highlights; no

shadows (occlusion of light sources)secondary illumination (reflection)transmission (transparency)focusing effects (refraction)

No obvious way to add these effects to “local” shading model


Ray Casting for Analytic Geometry Sampling

Address Visibility, Shading separatelyResolve visibility per-pixel using ray-primitive intersectionEvaluate local lighting model once at each visible pixel

eye

��

eye

p

n

p

n

ρ

ρ

light source

(no)

Use routine Shade() for lighting evaluation (Phong model):

// Shade obj surface at point ShadeP, normal ShadeN,

// as seen along direction Along

Radiance Shade ( Object obj, Ray Along,

Point ShadeP, Vector ShadeN ) {

... // sum Phong term for each light

}


Ray Casting with “Analytic” Visibility

Hit Cast ( Point R, Ray D ) {

find minimum t>0 such that R + t * D hits object

if ( object hit )

return ( object identifier, t, N )

else

return ( background object identifier, t )

} // Cast

eye

��

% ivray raycast.iv -e raycast.env -s L

FrameBuffer Render ( frustum, viewport ) {

Framebuffer FB[w][h];

For each raster y

For each pixel x

E = direction from eye through pixel x, y

hit = Cast ( eye, E );

FB[x][y] = Shade ( hit.obj, E, eye + hit.t * E, hit.N );

return FB

} // Render by Ray Casting


Ray Casting for Visibility

Advantages?Smooth variation of silhouettes; depth, normal, shading

Generality: can render anything withwhich a ray can be intersected !

Compactness of representationDisadvantages?

Time complexity (N objects, R pixels)Wasted work at “background” pixels


Ray Casting and Shadow Tests

Idea (Appel, 1968):Cast ray from eye through each pixelDetermine closest object along ray (and visible point there)Shade by summing contributions from unoccluded lights

Non-recursive! (But improved quality, realism)(Primary) shadows handled w/ existing capability!

eye

��

light

light

��

% ivray raycast.iv -e raycast.env -s S


Recursive Ray Tracing

Extend to reflection, refractionShading must take entire scene into accountRay tracing is a “global illumination algorithm”

Idea: Light originates at light sources, so“trace” photon paths from the light source

Known as Forward Ray Tracing:At each interaction, surface properties dictate

absorption, reemission, transmission probability

eye

O

light

��

Typically expressed as BRDF f (θi, φi, θe, φe) ∈ [0..1]Disadvantages of Forward Ray Tracing?


Forward Ray Tracing

Disadvantagesvery few of the photons end up at the eyevery hard to know in which directions

photons should be sentenormous number of cycles expended per photon

(can be ameliorated by packet tracing)result is usually objectionable noise


Backward Ray Tracing

Insight: we “see” only those rays that arrive at the eyeSo, trace “eye rays” E backward into scene

Find contributions to shading (only) at surface pointsEvaluate contributions along the viewing direction

eye

O

light

��

N

R

S

T

E

% ivray raycast.iv -e raycast.env -s R -d 1

Likely directions in which to sample:Shadow rays S (to light sources)Reflection rays R (along specular direction)Refraction rays T (along refraction direction)

Note: Shading operation is recursive !


Recursive Ray Tracing: Examples



% ivray sphereplane.iv -e raycast.env -s R -d 2

Experiment with reference ivray to understand these


Physical Units for Ray Tracing

From radiometry, measurement of EM energy (distinctfrom photometry, visual sensation of EM energy)

Radiance L:[power] / ([src area] [rcvr steradian])

“Power per unit projected area perpendicularto the ray per unit solid anglein the direction of the ray”

dA dω

A dA

POWER

AREA= RADIOSITYTOTAL EXITANT POWER

POWER

AREA STERADIAN= RADIANCE

SOURCERECEIVER

AS SEEN FROM SOURCE


Radiance Propagation

Consider two virtual spheres of radius r1, r2centered at differential source element dA, and

the patches R1,R2 defined on them by dω

dA

dω

R2

R1

Power flowing through R1 is P1, through R2 is P2

L1 =P1

A1 · dω1L2 =

P2

A2 · dω2

How are L1 and L2 related?


Radiance Propagation

Clearly P1 = P2; A1 = A2 = dA; dω1 = dω2

dA

dω

R2

R1

L1 =P1

dA · dω1=

P2

dA · dω2= L2

Radiance is constant along a ray!(What does this assume about propagation medium?)

Analogous derivation for fixed-size receiver


Response of a Sensor due to Radiance

Consider a small exposure meter whose field of viewimpinges on a large, uniformly illuminated surfaceSensors: retinal cells; film grains; CCD elements...

(after Hanrahan, 1993)

Aperture Sensor

What is total power impinging on sensor?Proportional to total surface area visible to sensorProportional to solid angle subtended by sensor

as viewed from the source(this is just the fraction of source’s energy received by the sensor)


Response of a Sensor due to Radiance

Once again, radial dependence cancels; conclude:Sensor response proportional to surface radiance!

(after Hanrahan, 1993)

Aperture Sensor

Thus, for two reasons:radiance constant along a raysensor response proportional to radiance

radiance is the quantity that shouldbe associated with a propagating ray!


Radiance and Reflection

Suppose a light source illuminates a surface patchConsider a viewing ray bouncing among perfect mirrors (ks = 1)

dA

dωL

E

Is there an equivalent situation with no mirrors?Now assign mirrors coefficients of reflection ks, kd < 1

What happens?


Reflection, Transmission Rays

Radiance is attenuated by reflection!

dA

dω

L

kS1

LkS1

kS2

LkS1 kS2

In these units, no attenuation due to distance from surface


Backward Ray Tracer Pseudo-Code

global int maxdepth = d; // trace only d bounces

RayTrace ( frustum, viewport ) {

For each raster y

For each pixel x

E = ray from eye through pixel x, y

FrameBuffer[x][y] = Trace ( eye, E, 1 )

} // RayTrace

Note: RayCast() returns hit structure containing:scene object hit->objectintersection parameter hit->tsurface point hit->P, normal hit->N

Radiance Trace ( Point R, Ray D,

int depth ) {

Hit *hit = RayCast ( R, D );

if ( hit )

return Shade ( hit->object, D, hit->P, hit->N, depth );

else

return Background ( R, D );

} // Trace

Note: Trace() returns a Radiance valueFor now, we’ll interpret this as an RGB valueLater in term, we’ll connect physical units to perceptual sensation






Point ShadeP, Vector ShadeN, int depth ) {

Radiance r = (0,0,0); // used to aggregate lighting terms

// collect ambient contribution

r += environment’s ambient radiance;

// collect local illumination contributions

For ( each light ) {

Ray sRay = Ray from point shadeP to light source;

if ( sRay . ShadeN > 0 ) {

r += diffuse contribution * light visibility

r += specular contribution * light visibility

}

}

// terminate recursion

if ( depth == maxdepth ) return r;

// continued

...



...

// Shade() continued

if ( object reflective ) {

Radiance rRefl;

Ray rRay = reflection ray direction;

rRefl = Trace ( ShadeP, rRay, depth + 1 )

r += rRefl * material specular coefficient

}

if ( object transparent ) {

Ray tRay = refraction ray direction;

if ( not total internal reflection ) {

Radiance rRefr;

rRefr = Trace ( ShadeP, tRay, depth + 1 )

r += rRefr * material transmission coefficient

}

else {

// ... TBD (Asst 6)

}

}

// return aggregate radiance

return r;

} // Shade


Revised Backward RT Pseudo-Code

Want to avoid spending many cycles for little radianceParameter maxdepth has no notion of radiance

Trace takes an additional parameter, weightand returns a Radiance value

We define a new global: float minweight

global int maxdepth = d; // trace only d bounces, or

global float minweight = w; // until attenuation exceeds w

RayTrace ( resolution, view frustum ) {

For each raster y

For each pixel x

E = ray from eye through pixel x, y

FrameBuffer[x][y] = Trace ( eye, E, 1, 1.0 )

} // RayTrace

We modify Trace() to take weight parameter:Radiance Trace ( Point R, Ray D,

int depth, float weight ) {

Hit *hit = RayCast ( (R,D) );

if ( hit )

return Shade ( hit->object, D, hit->P,

hit->N, depth, weight )

else

return Background ( R, D )

} // Trace


Revised Shade() Pseudo-Code




Point ShadeP, Vector ShadeN,

int depth, float weight ) {

... // same as previous pseudo-code

// conditionally spawn reflection ray

if ( obj->k_s * weight >= minweight ) {

Ray rRay = reflection ray

radiance += obj->k_s

* Trace ( ShadeP, rRay,

depth + 1, obj->k_s * weight )

}

// conditionally spawn transmission ray

if ( obj->k_t * weight >= minweight ) {

Ray tRay = refraction ray

if ( not total internal reflection )

radiance += obj->k_t

* Trace ( ShadeP, tRay,

depth + 1, obj->k_t * weight )

}

return radiance;

} // Shade


Example Call Stack of Trace, Shade

kS1

kS2

source

R

D

R’

N1D’

R’’

N2 D’’

Trace ( R, D, 1, 1.0 );

Shade ( object, D, R’, N1, 1, 1.0 ) // 1st surface

// radiance += k_s1 * Trace ( R’, D’, 2, k_s1 )

tmp = Trace ( R’, D’, 2, k_s1 )

Shade ( object, D’, R’’, N2, 2, k_s1 ) // 2nd surface

radiance += k_s2 * f ( D’, R’’, N2, src )

...

return radiance L; // k_s2 * ... src

radiance += k_s1 * L;

return radiance; // k_s2 * k_s1 * ... src

return radiance; // k_s2 * k_s1 * ... src


Ray Spawning / Termination

Termination conditions:maxdepth exceededminweight arises from multiple reflectionsRay leaves the scene (encounters background)

When might these criteria work poorly?Example problematic scenes?


Recap: Ray Tracer Components

Sample generatorOne eye-ray per pixel

(You will do super-sampling in Asst 6)Intersection finder

Workhorse function RayCast

Simple bounding-box test for accelerationShader, Secondary Ray generator

Radiance Aggregation (base case)Shadow (uses RayCast for shadow testing)ReflectionRefraction


Assignments 5 & 6 Infrastructure

Invoking ivray:% ivray [-d maxdepth] [-s shademode]

where shademode controls the shading function as:

[-s shademode (F,N,A,L,S,R)

Flat (use material base color)

Normal (use object normal as color)

Ambient (use ambient light only)

Local (use ambient, diffuse, & specular)

Shadows ("L" mode, with shadowing)

Recursive ("S" mode, with recursive rays)]

Other command-line arguments:Recursion termination conditionsSampling, anti-aliasing, etc.


Ray Tracing Extensions

Modify geometry (position or normal)Modify material characteristics (ray spawning)Simulate optical paths, camera geometry, wavelength distributionSample temporally to simulate motionAdaptive/hybrid sampling strategiesVisualizationAcceleration data structures (future lecture)


Bump Mapping

Perturb normals according to texture or function

(Stanford Graphics Lab)


Displacement Mapping

Perturb surface position; account for shadows, interreflection etc.

(Utah Graphics Lab)


Glossy Materials

Displace origin of reflection, transmission rays

(Andy Whittock, Toonnation)


Soft Shadows

Subdivide and sample area light sources

(Cornell Computer Graphics Lab)


Optical Effects

Simulate camera optical path for various effects:Depth of field; barrel distortion; vignetting; etc.

Also, wavelength-dependent effects (e.g., chromatic aberration)(Kolb & Hanrahan, Siggraph ’95)


Caustics

Identify extremal paths from source to surfaces

(Hanrahan & Mitchell, Siggraph ’92)


Motion Blur

Simulate finite (non-zero) shutter speed; sample temporally


Adaptive Sampling (Gridded)

Idea: Show coarse picture quicklySample along coarse grid, interpolate, refine


Adaptive Sampling (Radiance-Based)

Idea: Hybrid of ray-tracing, screen-space interpolationSubdivide whenever radiance is discontinuous, or has high gradient

(Research of graphics lab alumna, Kavita Bala, 1990)


Visualization

Show scene, camera, lights, ray intersections etc.Example: Ben Garlick’s SGI demo “flyray”:cd /mit/6.837/demos/flyray/data; ../flyray

(Parallel application)


Next week:

Tuesday: Animation & SimulationThursday: Final Project Brainstorming


Secondary Rays: Reflection

p

LN

RL

L

L

−L

R

Compute reflection ray R as:

L‖ = N(L · N)

R = L‖ − L⊥= N(L · N) − (L − N(L · N))

= N(L · N) − L + N(L · N)

= 2N(L · N) − L


Secondary Rays: Refraction

All media have index of refraction ηratio of c (in vacuum) to light speed (in material)Of course, η ≥ 1 for all physical materials

��

N

−N

I

θi

θt

T

M

ηt

ηi

ηt > ηi

sparse

dense

Consider boundary admitting incident (i)and transmitted (t) rays

Snell’s law says that, at this boundary

sin θi

sin θt=

ηt

ηior

ηi sin θi = ηt sin θt

(actually, η depends on λ, causing dispersion)


Computing Refraction Vector

Plugging it all in:

��

N

−N

I

θi

θt

T

M

ηt

ηi

ηt > ηi

T

T

I

I

T = −ηi

ηtI + N

ηi

ηtcos θi −

√√√√√√√√1 − (ηi

ηt)2(1 − cos2 θi)

Eliminating cosines:

T = −ηi

ηtI + N

ηi

ηtN · I −

√√√√√√√√1 − (ηi

ηt)2(1 − (N · I)2)

What if expression under radical is negative?


assignment 5 (ray casting) extensiongroups.csail.mit.edu/graphics/classes/6.837/f02/... ·...

Documents