Reflected-Scene Impostors for Realistic Reflections at Interactive - - PowerPoint PPT Presentation

reflected scene impostors for realistic reflections at
SMART_READER_LITE
LIVE PREVIEW

Reflected-Scene Impostors for Realistic Reflections at Interactive - - PowerPoint PPT Presentation

Voicu Popescu, Chunhui Mei, Jordan Dauble, and Elisha Sacks Purdue University Reflected-Scene Impostors for Realistic Reflections at Interactive Rates Reflectionsa difficult problem Every reflector is a portal onto a world which is as


slide-1
SLIDE 1

Reflected-Scene Impostors for Realistic Reflections at Interactive Rates

Voicu Popescu, Chunhui Mei, Jordan Dauble, and Elisha Sacks Purdue University

slide-2
SLIDE 2

2

Reflections—a difficult problem

  • Every reflector is a portal onto a world

which is as rich as the directly observed scene and which has complex image formation laws

slide-3
SLIDE 3

3

Prior work—vast

Approximation of reflected scene Feed-forward reflection rendering Image-Based Rendering Ray tracing

slide-4
SLIDE 4

4

Problem of rendering reflections

  • Compute

– Intersection with reflector – Reflected ray – Intersection with reflected scene – antialiasing

slide-5
SLIDE 5

5

Problem of rendering reflections

  • Compute

– Intersection with reflector – Reflected ray – Intersection with reflected scene – antialiasing

“OpenGL” ???

slide-6
SLIDE 6

6

Reflected-scene approximation

  • Reflected scene replaced with approx. that

provides

– Fast intersection with ray – Antialiasing

slide-7
SLIDE 7

7

Reflected-scene approximation

  • Example: environment mapped reflections

– Reflected scene infinitely far away – Straight forward intersection with ray – Antialiasing computed in 2D (mipmapping)

slide-8
SLIDE 8

8

Reflected-scene approximation

  • Example: environment mapped reflections

– Reflected scene infinitely far away – Straight forward intersection with ray – Antialiasing computed in 2D (mipmapping) – Drastic approximation, incorrect results close to the reflector

slide-9
SLIDE 9

9

Our approach

  • Approximate reflected scene with

impostors

– Considerable prior work on impostors – Reflector surface prevents desired viewpoint from getting too close to the impostor – Reflection distortion hides impostor artifacts

slide-10
SLIDE 10

10

Impostor requirements

  • Impostor has to provide

– Fast construction – Fast intersection with ray – Antialiasing

slide-11
SLIDE 11

11

Results: billboard impostors

slide-12
SLIDE 12

12

Results: depth image impostors

slide-13
SLIDE 13

13

Billboard impostors

  • Replace reflected object with billboard
  • Higher order reflections

– Reflective billboards (normal mapped quads)

slide-14
SLIDE 14

14

Billboard impostors

  • Impostor has to provide

– Fast construction YES – Fast intersection with ray YES – Antialiasing YES

slide-15
SLIDE 15

15

Pixel algorithm

  • For D diffuse, R reflective billboards, and

maximum reflection order K

– Compute reflected ray r – For reflection order 1 to K

  • Intersect with (D+R-1) billboards
  • If no intersection

– return EM(r)

  • Else if intersection with diffuse billboard DBi

– return DBi(r)

  • Else if intersection with reflective billboard DBi

– r = DBi(r)

slide-16
SLIDE 16

16

Pixel algorithm

  • For D diffuse, R reflective billboards, and

maximum reflection order K

– Compute reflected ray r – For reflection order 1 to K

  • Intersect with (D+R-1) billboards
  • If no intersection

– return EM(r)

  • Else if intersection with diffuse billboard DBi

– return DBi(r)

  • Else if intersection with reflective billboard DBi

– r = DBi(r)

O(K*(D+R))

slide-17
SLIDE 17

17

Example: 4 teapots

  • D = 1, R = 4,

D+(R-1)+D = 5 intersections / pix

  • 12 second order

reflections

  • 40fps
slide-18
SLIDE 18

18

Example: table scene

  • D = 2, R = 2,

D+(R-1)+D = 5 intersections / pix

  • 2 second order

reflections

  • 33 fps
slide-19
SLIDE 19

19

Example: table scene

  • D = 2, R = 2,

D+(R-1)+D = 5 intersections / pix

  • 2 second order

reflections

  • 33 fps
slide-20
SLIDE 20

20

Example: table scene

slide-21
SLIDE 21

21

Example: pushing-it scene

  • D = 2, R = 9,

D+(R-1)+D = 11 intersections / pix

  • 72 second order

reflections

  • 11 fps
slide-22
SLIDE 22

22

Example: pushing-it scene

  • D = 2, R = 9,

D+(R-1)+D = 11 intersections / pix

  • 72 second order

reflections

  • 6 fps
slide-23
SLIDE 23

23

Example: pushing-it scene

slide-24
SLIDE 24

24

Problem

Transition from impostor to environment map (red in left image) is discontinuous.

slide-25
SLIDE 25

25

Solution: ray morphing

slide-26
SLIDE 26

26

Solution: ray morphing

E R r0 re r1 B a r1

d

ra rm rm=ra+(r1

d-ra)h/H

A q0 q1

slide-27
SLIDE 27

27

Solution

Left—continuous transition. Right—morph region (green), environment map (red).

slide-28
SLIDE 28

28

Ray morphing

slide-29
SLIDE 29

29

Attenuation w/ distance

slide-30
SLIDE 30

30

Fresnel

slide-31
SLIDE 31

31

Combined effects

slide-32
SLIDE 32

32

Animation and materials

slide-33
SLIDE 33

33

Comparison to env. mapping

slide-34
SLIDE 34

34

Billboard limitations

  • No support for objects very close to the

reflector

  • Limited accuracy

– Flat reflection – Lack of motion parallax

slide-35
SLIDE 35

35

Depth image impostors

  • Impostor has to provide

– Fast construction YES – Fast intersection with ray ??? – Antialiasing YES

slide-36
SLIDE 36

36

Depth image—ray intersection

Epipolar-like constraints: intersection computed as 1D search Still too many steps along epipolar segment

slide-37
SLIDE 37

37

Simplified Rotated Depth Maps

Pre-rotate depth map. All rays ever needed project to rows. Pre-simplify rows.

slide-38
SLIDE 38

38

Simplified Rotated Depth Maps

slide-39
SLIDE 39

39

SRDM construction cost

980 480 300 210 Construction time [ms] 64 32 16 8 Number of segments

Rigid body transformations, color updates, and reflector updates do not require reconstruction.

slide-40
SLIDE 40

40

Depth image impostor results

slide-41
SLIDE 41

41

Depth image impostor results

slide-42
SLIDE 42

42

Depth image impostor results

slide-43
SLIDE 43

43

Depth image impostor results

slide-44
SLIDE 44

44

Depth image impostor results

slide-45
SLIDE 45

45

Depth image impostor results

slide-46
SLIDE 46

46

SRDM under-sampling

One rotated depth map every 20o, 10o, 3o, and 2o, respectively.

slide-47
SLIDE 47

47

Depth image impostor results

slide-48
SLIDE 48

48

Conclusions

  • The reflected-impostor approach works

– Fast, realistic – Increased modeling effort

  • Rendering reflections reduced to the

lesser problem of rendering w/ impostors

slide-49
SLIDE 49

49

Future work

  • Other types of impostors

– occlusion-resistant

slide-50
SLIDE 50

50

Future work

  • Other types of impostors
  • Other BRDFs
  • Self-reflections
  • Constructing the SRDMs on the GPU
slide-51
SLIDE 51

51

Acknowledgments

  • Funding & equipment

– NSF, Intel, Microsoft, Computer Science Purdue, Visualization Laboratory Purdue

  • Stanford 3D Scanning
  • Rep. for models
  • Paul Debevec for

environment maps

  • Our graphics group at

Purdue for miscellaneous but important help