Efficient minimal-surface regularization of perspective depth maps in variational stereo

Graber, Gottfried; Balzer, Jonathan; Soatto, Stefano; Pock, Thomas

doi:10.1109/cvpr.2015.7298649

Cited by 34 publications

(33 citation statements)

References 41 publications

(46 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Samples {I t } can be used to compute an approximation of ρ, S, for instance in the sense of maximum-likelihood, with suitable regularization [13,15] ρ,Ŝ = arg max ρ,S,gt p({I t }|ρ, S) + λR(S)…”

Section: Point-estimate Approximation: R-hogmentioning

confidence: 99%

Multi-view feature engineering and learning

Dong

Karianakis

Davis

et al. 2015

2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

Self Cite

View full text Add to dashboard Cite

Section: Point-estimate Approximation: R-hogmentioning

confidence: 99%

Multi-view feature engineering and learning

Dong

Karianakis

Davis

et al. 2015

2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

Self Cite

View full text Add to dashboard Cite

“…where f is the focal length of the polarisation camera in the x and y directions and (x 0 , y 0 ) is the principal point. The direction of the outward pointing surface normal is defined as the cross product of the partial derivatives with respect to x and y [11]:…”

Section: Perspective Depth Representationmentioning

confidence: 99%

“…We now use the surface normals estimated by the graphical model optimisation to compute an albedo map. In principal, the albedo can be computed from these normals and the unpolarised intensity simply by rearranging (11). However, this purely local estimation is unstable and noise in the normals leads to artefacts in the estimated albedo map.…”

Section: Albedo Estimation With Gradient Consistencymentioning

confidence: 99%

Depth From a Polarisation + RGB Stereo Pair

Zhu

Smith

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Estimated specular mask Corrected normal Depth from stereo (a) Setup and Input (b) Local disambiguation Estimated Albedo Estimated Shape (c) Output Section 4 Section 5 Section 6 Figure 1: Overview: From a stereo pair of one polarisation image and one RGB image (a) we merge stereo depth with polarisation normals using a higher order graphical model (b) before estimating an albedo map and the final geometry (c). AbstractIn this paper, we propose a hybrid depth imaging system in which a polarisation camera is augmented by a second image from a standard digital camera. For this modest increase in equipment complexity over conventional shapefrom-polarisation, we obtain a number of benefits that enable us to overcome longstanding problems with the polarisation shape cue. The stereo cue provides a depth map which, although coarse, is metrically accurate. This is used as a guide surface for disambiguation of the polarisation surface normal estimates using a higher order graphical model. In turn, these are used to estimate diffuse albedo. By extending a previous shape-from-polarisation method to the perspective case, we show how to compute dense, detailed maps of absolute depth, while retaining a linear formulation. We show that our hybrid method is able to recover dense 3D geometry that is superior to state-of-the-art shape-from-polarisation or two view stereo alone.

show abstract

“…However, the per-pixel minimum might not necessarily be accurate or lead to smooth surface due to factors such as specular reflection, self occlusion and intensity ambiguity in texture-less areas. In order to solve this, previous methods [26,17,16] have employed a total variational minimization to remove noise. The goal is to minimize the gradient of depth map D to produce smooth surface and preserve depth discontinuity around edges, which is achieved via minimizing…”

Section: Depth Estimationmentioning

confidence: 99%

Realtime Dynamic 3D Facial Reconstruction for Monocular Video In-the-Wild

Liu

Wang

Yang

et al. 2017

2017 IEEE International Conference on Computer Vision Workshops (ICCVW)

View full text Add to dashboard Cite

With the increasing amount of videos recorded using 2D mobile cameras, the technique for recovering the 3D dynamic facial models from these monocular videos has become a necessity for many image and video editing applications. While methods based parametric 3D facial models can reconstruct the 3D shape in dynamic environment, large structural changes are ignored. Structure-frommotion methods can reconstruct these changes but assume the object to be static. To address this problem we present a novel method for realtime dynamic 3D facial tracking and reconstruction from videos captured in uncontrolled environments. Our method can track the deforming facial geometry and reconstruct external objects that protrude from the face such as glasses and hair. It also allows users to move around, perform facial expressions freely without degrading the reconstruction quality.

show abstract

Efficient minimal-surface regularization of perspective depth maps in variational stereo

Cited by 34 publications

References 41 publications

Multi-view feature engineering and learning

Multi-view feature engineering and learning

Depth From a Polarisation + RGB Stereo Pair

Realtime Dynamic 3D Facial Reconstruction for Monocular Video In-the-Wild

Contact Info

Product

Resources

About