Efficient Surface Reconstruction using Generalized Coulomb Potentials
read more
Citations
Point Cloud Skeletons via Laplacian Based Contraction
On surface reconstruction: A priority driven approach
A two-level approach to implicit surface modeling with compactly supported radial basis functions
Surface Reconstruction Based on Hierarchical Floating Radial Basis Functions
Continuous global optimization in surface reconstruction from an oriented point cloud
References
Fronts propagating with curvature-dependent speed: algorithms based on Hamilton-Jacobi formulations
A hierarchical O(N log N) force-calculation algorithm
A volumetric method for building complex models from range images
Surface reconstruction from unorganized points
Poisson surface reconstruction
Related Papers (5)
Frequently Asked Questions (19)
Q2. What are the future works mentioned in the paper "Efficient surface reconstruction using generalized coulomb potentials" ?
Further work concerns more efficient smoothing of implicit surfaces on non-uniform grids based on curvature flows and/or ( anisotropic ) diffusion, and improving the overall performance of the polygonization method. Of particular interest is the possibility of accelerating the computations using modern, programmable GPU hardware.
Q3. What is the force field in which the surface is convected?
The force field in which the surface is convected is based on generalized Coulomb potentials evaluated on an adaptive grid (i.e., an octree) using the fast, hierarchical algorithm of Barnes and Hut [6].
Q4. How many voxels would be required to reconstruct the Lucy statue?
Reconstructing these models with methods such as the FFT [19] or the method in [18] would require two uniform grids of size 20483 voxels, i.e., more than 60 GB of memory.
Q5. How can the authors use the adaptive grid to evaluate Coulomb potentials?
since accurate representations are only required close to the data set, the authors can use an adaptive grid based on an octree to efficiently evaluate Coulomb potentials and to represent and triangulate the implicit function.
Q6. What are the main reasons why Morse and Ohtake use compactly supported RBFs?
Morse et al. [23] and Ohtake et al. [25] use compactly-supported RBFs to achieve local control and reduce computational costs by solving a sparse linear system.
Q7. What is the traditional approach to reconstructed surfaces?
The traditional approach is to compute a signed distance function and represent the reconstructed implicit surface by an iso-contour of this function [5,8,12,17].
Q8. What is the complexity of the polygonization step?
considering that a constant number of cells are visited for evaluating χ̃ at a point (see Algorithm 4), the complexity of the polygonization step is linear in the total number of nodes.
Q9. What is the recent proposal by Ohtake et al.?
Ohtake et al. [24] proposed the so-called ’partition of the unity implicits’, which can be regarded as the combination of algebraic patches and RBFs.
Q10. What is the noise in the data?
Since this data set is contaminated by noise, interpolating methods such as the Power Crust generate very noisy surfaces with holes due to the non-uniformity of the samples.
Q11. Does the FFT method cope well with shot noise?
Although the FFT method is resilient to Gaussian noise, it does not cope well with shot noise, as this affects the whole Fourier spectrum and not only certain (high) frequencies.
Q12. What is the complexity of the marching procedure?
Since each interior cell is visited at most once during convection, the complexity of the marching procedure is O(M · logM), where logM comes from the heap sort algorithm.
Q13. What is the complexity of the fast convection algorithm?
Since the complexity of evaluating the Coulomb potential (see Algorithm 2) at the center of a grid cell is O(logM), the total complexity of the fast convection algorithm is O(M · log2 M).
Q14. How does the fast convection algorithm work?
This is required by the fast convection algorithm (described in Section 4.3) which, starting from the bounding box of the grid, follows increasing paths of the scalar field until regional maxima (corresponding to the sample points) and ridges are reached.
Q15. What is the number of triangles of the reconstructed surface?
as their polygonizer is based on tetrahedral decomposition, the number of triangles of the reconstructed surface is quite high.
Q16. What is the average computation time of the Poisson method?
The computation time of their method is comparable to that of the MPU method, which is one of the fastest geometrically-adaptive reconstruction methods according to [20, 24], while the peak memory usage of their method (at octree depth D = 9 with m = 5) remains well below those of the other methods.
Q17. What is the main reason why Zhao et al. developed a 'geo?
The work of Zhao et al. [31] inspired researchers from the computational-geometry community to develop ’geometric convection’ algorithms [2, 11] in the context of surface reconstruction.
Q18. What is the maximum tree depth for a non-empty leaf?
When the maximum tree depth D is reached for a non-empty leaf l, l ∈ O, the node is not further subdivided if a new particle is to be assigned to it.
Q19. How did Giesen and John introduce the notion of flow in computational geometry?
Giesen and John [16] introduced the notion of flow in computational geometry, and Scheidegger et al. [28] proposed an adaptive method based on the Moving Least-Squares algorithm.