image-based bidirectional scene reprojectionhhoppe.com/bireproj_supp.pdf · 2019. 5. 9. ·...

3
Image-Based Bidirectional Scene Reprojection Supplemental Material Lei Yang 1 Yu-Chiu Tse 1 Pedro V. Sander 1 Jason Lawrence 2 Diego Nehab 3,4 Hugues Hoppe 3 Clara L. Wilkins 5 1 Hong Kong UST 2 University of Virginia 3 Microsoft Research 4 IMPA 5 Wesleyan University Supplemental results This document contains expanded figures with supplemental results for each of the applications presented in the paper. More specifically, Figure 1–4 include the rendering results and difference images for all B-frames, and MSE error graphs for all techniques. Comparison to frame upsampling Figure 5 shows example B-frame renderings of our approach compared to that of Didyk et al. [2010]. We must stress that this is not a completely fair comparison, since the method of Didyk et al. [2010] is targeted at higher-frame-rate displays, where the individual B-frames are hardly perceived. We included this comparison to demonstrate that our image-based bidirectional reprojection technique is the first real-time technique that interpolates frames in image-space with sufficient quality for rendering on standard displays with common refresh rates. Linear Blending Single-dir. reproj. (no reshade) Scene-assisted bidir. reproj. Single-dir. reproj. (reshade on miss) Reference Image-based bidir. reproj. I t I t+1 I t+0.25 I t+0.5 I t+0.75 0% 5% 10% 0 10 20 30 40 50 60 70 80 4.3 4.35 4.4 4.45 4.5 4.55 4.6 Render Time (msec) Animation time (sec) Performance 0 0.005 0.01 0.015 0.02 0.025 0.03 4.3 4.35 4.4 4.45 4.5 4.55 4.6 MSE [0,1] Animation time (sec) Quality (MSE) 30 40 50 60 70 80 90 100 4.3 4.35 4.4 4.45 4.5 4.55 4.6 SSIM (%) Animation time (sec) Quality (SSIM) Figure 1: Results of our algorithm on the walking scene. The lines in the plots are colored according to the color of the frame around the insets of the corresponding method. The closeups show error images.

Upload: others

Post on 18-Aug-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Image-Based Bidirectional Scene Reprojectionhhoppe.com/bireproj_supp.pdf · 2019. 5. 9. · Image-Based Bidirectional Scene Reprojection Supplemental Material Lei Yang 1Yu-Chiu Tse

Image-Based Bidirectional Scene ReprojectionSupplemental Material

Lei Yang1 Yu-Chiu Tse1 Pedro V. Sander1 Jason Lawrence2 Diego Nehab3,4 Hugues Hoppe3 Clara L. Wilkins5

1Hong Kong UST 2University of Virginia 3Microsoft Research 4IMPA 5Wesleyan University

Supplemental resultsThis document contains expanded figures with supplemental results for each of the applications presented in the paper. Morespecifically, Figure 1–4 include the rendering results and difference images for all B-frames, and MSE error graphs for alltechniques.

Comparison to frame upsamplingFigure 5 shows example B-frame renderings of our approach compared to that of Didyk et al. [2010]. We must stress that this isnot a completely fair comparison, since the method of Didyk et al. [2010] is targeted at higher-frame-rate displays, where theindividual B-frames are hardly perceived. We included this comparison to demonstrate that our image-based bidirectionalreprojection technique is the first real-time technique that interpolates frames in image-space with sufficient quality for renderingon standard displays with common refresh rates.

Linear Blending Single-dir. reproj. (no reshade)

Scene-assisted bidir. reproj. Single-dir. reproj. (reshade on miss)

ReferenceImage-based bidir. reproj.

ItIt+1It+0.25

It+0.5 It+0.75

0% 5% 10%

0

10

20

30

40

50

60

70

80

4.3 4.35 4.4 4.45 4.5 4.55 4.6

Ren

der

Tim

e (m

sec)

Animation time (sec)

Performance

0

0.005

0.01

0.015

0.02

0.025

0.03

4.3 4.35 4.4 4.45 4.5 4.55 4.6

MS

E [0

,1]

Animation time (sec)

Quality (MSE)

30

40

50

60

70

80

90

100

4.3 4.35 4.4 4.45 4.5 4.55 4.6

SS

IM (

%)

Animation time (sec)

Quality (SSIM)

Figure 1: Results of our algorithm on the walking scene. The lines in the plots are colored according to the color of the frame around theinsets of the corresponding method. The closeups show error images.

Page 2: Image-Based Bidirectional Scene Reprojectionhhoppe.com/bireproj_supp.pdf · 2019. 5. 9. · Image-Based Bidirectional Scene Reprojection Supplemental Material Lei Yang 1Yu-Chiu Tse

Single-dir. reproj. (no reshade)

Scene-assisted bidir. reproj. Image-based bidir. reproj. ReferenceSingle-dir. reproj. (reshade on miss)

0

100

200

300

400

500

600

700

800

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

Ren

der

Tim

e (m

sec)

Animation time (sec)

Performance

0

0.0005

0.001

0.0015

0.002

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

MS

E [0

,1]

Animation time (sec)

Quality (MSE)

88

90

92

94

96

98

100

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

SS

IM (

%)

Animation time (sec)

Quality (SSIM)

Figure 2: Results of our algorithm on the walking scene with motion blur.

Linear Blending Single-dir. reproj. (no reshade)

Scene-assisted bidir. reproj. Single-dir. reproj. (reshade on miss)

ReferenceImage-based bidir. reproj.

ItIt+1It+0.25

It+0.5 It+0.75

0% 5% 10%

0

20

40

60

80

100

120

140

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

Ren

der

Tim

e (m

sec)

Animation time (sec)

Performance

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

MS

E [0

,1]

Animation time (sec)

Quality (MSE)

30

40

50

60

70

80

90

100

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

SS

IM (

%)

Animation time (sec)

Quality (SSIM)

Figure 3: Results of our algorithm on the terrain scene. The lines in the plots are colored according to the color of the frame around the insetsof the corresponding method.

Page 3: Image-Based Bidirectional Scene Reprojectionhhoppe.com/bireproj_supp.pdf · 2019. 5. 9. · Image-Based Bidirectional Scene Reprojection Supplemental Material Lei Yang 1Yu-Chiu Tse

Linear Blending Single-dir. reproj. (no reshade)

Scene-assisted bidir. reproj.

Reference

Image-based bidir. reproj.

ItIt+1It+0.25

It+0.5 It+0.75

0% 5% 10%

0

5

10

15

20

25

30

35

4.11 4.14 4.17 4.2 4.23 4.26

Ren

der

Tim

e (m

sec)

Animation time (sec)

Performance

0

0.0005

0.001

0.0015

0.002

0.0025

0.003

0.0035

0.004

4.11 4.14 4.17 4.2 4.23 4.26

MS

E [0

,1]

Animation time (sec)

Quality (MSE)

80

85

90

95

100

4.11 4.14 4.17 4.2 4.23 4.26

SS

IM (

%)

Animation time (sec)

Quality (SSIM)

Figure 4: Results of our algorithm on the NVIDIA human head scene. The lines in the plots are colored according to the color of the framearound the insets of the corresponding method.

Image-based bidirectional reprojection

Frame upsampling for high-refresh-rate displays [Didyk et al. 2010]

It (original) It+1 (original)It It+0.33 It+0.67 It+1

Figure 5: Comparison with the method of Didyk et al. [2010], which is suitable for high-frame-rate displays where individual interpolatedframes are hardly visible. (Note that I-frames are not identical to ours because Didyk et al. [2010] use contrast-enhanced images to counteractblurriness in the interpolated frames.)