1
[ 1] Tadano, R., Kumar Pediredla, A. and Veeraraghavan, A., Depth selective camera: A direct, on-chip, programmable technique for depth selectivity in photography. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), (pp. 3595-3603). [2] Pediredla A., Dave A., and Veeraraghavan A. 2019. SNLOS: Non-line-of-sight Scanning through Temporal Focusing. 2019 IEEE International Conference on Computational Photography (ICCP). Adithya Kumar Pediredla 1,2 , Ashok Veeraraghavan 1 , Ioannis Gkioulekas 2 1 Rice University, 2 Carnegie Mellon University Ellipsoidal Path Connections for Time-Gated Rendering https://github.com/ cmu-ci-lab/MitsubaToFRenderer Launch-and-play image file for Amazon Web Services clouds (ami-28308957). Trivially scales on multiple cores, CPUs, cloud clusters Time = 0.4s Time = 1.34s Time = 1.74s Steady state Existing BDPT Ours Gate location 16 ns Gate location 16 ns Gate location 16 ns Proximity detection camera Time = 1s Time = 5s Time = 5.85 s Steady state Existing BDPT Ours Gate location 4 ns Gate location 20 ns Gate location 23.4 ns Transients for dynamic scenes Path integral for physically-based rendering Steady state rendering: I j =: d(ത ), σ = Path, set of ordered vertices Radiance of the path, includes photometric and geometric light loss Path Measure (area or volumetric) Probability of path that depends on sampling strategy and the measure () Time-of-flight (TOF) renderer ellipsoidal vertex Ellipsoidal connections for efficient rendering LIDAR Amplitude modulated continuous wave 5 cm Ours Existing BDPT Scene SNLOS: Non-line-of-sight Scanning through Temporal Focusing [3] CWAM-ToF with Depth Selective Codes [2] Ours time 0 W t MSE = 0.056 MSE = 0.010 Existing BDPT References Acknowledgments: Gated Cameras Transient Cameras TOF RENDERING TASKS THEORY TIME-GATED RENDERING SCENARIOS PUBLICLY AVAILABLE SOURCE CODE RENDERING CUSTOM TOF CAMERAS Unified ToF Renderer: () ≔ |ത | d ; : path length LIDAR: = (|ത | − min I t >0 ) AMCW: = , |ത | ; : cross-correlation Gated/Transient Camera: = rect −|ത | Pulsed laser Camera Frame: 80 Frame: 150 Frame: 115 MSE = 0.411 MSE = 0.044 MSE = 0.183 MSE = 0.019 MSE = 0.442 MSE = 0.162

Ellipsoidal Path Connections for Time-Gated Rendering...Path integral for physically-based rendering Transients for dynamic scenes Steady state rendering: Ij=: 𝓅 𝜑𝐱തd𝜇(𝐱ത),

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Ellipsoidal Path Connections for Time-Gated Rendering...Path integral for physically-based rendering Transients for dynamic scenes Steady state rendering: Ij=: 𝓅 𝜑𝐱തd𝜇(𝐱ത),

[1] Tadano, R., Kumar Pediredla, A. and Veeraraghavan, A., Depth selective camera: A direct,on-chip, programmable technique for depth selectivity in photography. In Proceedings of theIEEE International Conference on Computer Vision (ICCV), (pp. 3595-3603).

[2] Pediredla A., Dave A., and Veeraraghavan A. 2019. SNLOS: Non-line-of-sight Scanningthrough Temporal Focusing. 2019 IEEE International Conference on Computational Photography(ICCP).

Adithya Kumar Pediredla1,2, Ashok Veeraraghavan1 , Ioannis Gkioulekas2

1Rice University, 2Carnegie Mellon University

Ellipsoidal Path Connections for Time-Gated Rendering

https://github.com/

cmu-ci-lab/MitsubaToFRenderer

• Launch-and-play image file for Amazon Web Services clouds (ami-28308957).

• Trivially scales on multiple cores, CPUs, cloud clusters

Time = 0.4s Time = 1.34s Time = 1.74s

Stea

dy

stat

eEx

isti

ng

BD

PT

Ou

rs

Gate location 16 ns

Gate location 16 ns

Gate location 16 ns

Proximity detection camera

Time = 1s Time = 5s Time = 5.85 s

Stea

dy

stat

eEx

isti

ng

BD

PT

Ou

rs

Gate location 4 ns

Gate location 20 ns

Gate location23.4 ns

Transients for dynamic scenesPath integral for physically-based rendering

Steady state rendering: Ij =: 𝓅𝜑 ത𝐱 d𝜇(ത𝐱),

≈𝟏

𝒏σ𝒊=𝟏𝒏 𝜑 ത𝐱𝒊

𝒑 ത𝐱𝒊

ത𝐱 → Path, set of ordered vertices

𝜑 → Radiance of the path, includes photometric and geometric light loss

𝜇 → Path Measure (area or volumetric)

𝒑 ത𝐱𝒊 → Probability of path ത𝐱𝒊 that depends on sampling strategy and the measure (𝝁)

Time-of-flight (TOF) renderer

ellipsoidal vertex𝑥𝑒

𝑥𝑠 𝑥𝑐

Ellipsoidal connections for efficient rendering

• LIDAR

• Amplitude modulated continuous wave

5 cm

OursExisting BDPTScene

SNLOS: Non-line-of-sight Scanning through Temporal Focusing [3]

CWAM-ToF with Depth Selective Codes [2]

Ours

time0

Wt

MSE = 0.056 MSE = 0.010

Existing BDPT

References

Acknowledgments:

• Gated Cameras

• Transient Cameras

TOF RENDERING TASKS THEORY TIME-GATED RENDERING SCENARIOS PUBLICLY AVAILABLE SOURCE CODE

RENDERING CUSTOM TOF CAMERAS

Unified ToF Renderer:𝐼𝑗(𝑓) ≔ 𝓅 𝑓 |ത𝐱| 𝜑 ത𝐱 d𝜇 ത𝐱 ; ത𝐱 : path length

• LIDAR: 𝑓 = 𝛿(|ത𝐱| − minI t >0

𝑡)

• AMCW: 𝑓 = 𝐶𝜔,𝜙 |ത𝐱| ; 𝐶: cross-correlation

• Gated/Transient Camera: 𝑓 = rect𝜏−|ത𝐱|

𝑌

𝑥𝑠 𝑥𝑐Pulsed laser

Camera

Frame: 80

Frame: 150Frame: 115 MSE = 0.411

MSE = 0.044

MSE = 0.183

MSE = 0.019

MSE = 0.442

MSE = 0.162