@inproceedings{erler-2020-p2s, title = "Points2Surf: Learning Implicit Surfaces from Point Clouds", author = "Philipp Erler and Paul Guerrero and Stefan Ohrhallinger and Michael Wimmer and Niloy Mitra", year = "2020", abstract = "A key step in any scanning-based asset creation workflow is to convert unordered point clouds to a surface. Classical methods (e.g., Poisson reconstruction) start to degrade in the presence of noisy and partial scans. Hence, deep learning based methods have recently been proposed to produce complete surfaces, even from partial scans. However, such data-driven methods struggle to generalize to new shapes with large geometric and topological variations. We present Points2Surf, a novel patch-based learning framework that produces accurate surfaces directly from raw scans without normals. Learning a prior over a combination of detailed local patches and coarse global information improves generalization performance and reconstruction accuracy. Our extensive comparison on both synthetic and real data demonstrates a clear advantage of our method over state-of-the-art alternatives on previously unseen classes (on average, Points2Surf brings down reconstruction error by 30% over SPR and by 270%+ over deep learning based SotA methods) at the cost of longer computation times and a slight increase in small-scale topological noise in some cases. Our source code, pre-trained model, and dataset are available on: https://github.com/ErlerPhilipp/points2surf ", month = oct, isbn = "978-3-030-58558-7", series = "Lecture Notes in Computer Science", publisher = "Springer International Publishing", location = "Glasgow, UK (online)", address = "Cham", event = "ECCV 2020", editor = "Vedaldi, Andrea and Bischof, Horst and Brox, Thomas and Frahm, Jan-Michael", doi = "10.1007/978-3-030-58558-7_7", booktitle = "Computer Vision -- ECCV 2020", journal = "Computer Vision – ECCV 2020", pages = "17", volume = "12350", pages = "108--124", keywords = "surface reconstruction, implicit surfaces, point clouds, patch-based, local and global, deep learning, generalization", URL = "https://www.cg.tuwien.ac.at/research/publications/2020/erler-2020-p2s/", } @article{ohrhallinger_stefan-2018-cgf, title = "FitConnect: Connecting Noisy 2D Samples by Fitted Neighborhoods", author = "Stefan Ohrhallinger and Michael Wimmer", year = "2019", abstract = "We propose a parameter-free method to recover manifold connectivity in unstructured 2D point clouds with high noise in terms of the local feature size. This enables us to capture the features which emerge out of the noise. To achieve this, we extend the reconstruction algorithm HNN-Crust, which connects samples to two (noise-free) neighbors and has been proven to output a manifold for a relaxed sampling condition. Applying this condition to noisy samples by projecting their k-nearest neighborhoods onto local circular fits leads to multiple candidate neighbor pairs and thus makes connecting them consistently an NP-hard problem. To solve this efficiently, we design an algorithm that searches that solution space iteratively on different scales of k. It achieves linear time complexity in terms of point count plus quadratic time in the size of noise clusters. Our algorithm FitConnect extends HNN-Crust seamlessly to connect both samples with and without noise, performs as local as the recovered features and can output multiple open or closed piece-wise curves. Incidentally, our method simplifies the output geometry by eliminating all but a representative point from noisy clusters. Since local neighborhood fits overlap consistently, the resulting connectivity represents an ordering of the samples along a manifold. This permits us to simply blend the local fits for denoising with the locally estimated noise extent. Aside from applications like reconstructing silhouettes of noisy sensed data, this lays important groundwork to improve surface reconstruction in 3D. Our open-source algorithm is available online.", month = feb, journal = "Computer Graphics Forum", volume = "38", number = "1", issn = "1467-8659", doi = "10.1111/cgf.13395", pages = "126--137", keywords = "curve fitting, noisy samples, guarantees, curve reconstruction", URL = "https://www.cg.tuwien.ac.at/research/publications/2019/ohrhallinger_stefan-2018-cgf/", } @inproceedings{ohrhallinger_stefan-2018-pg, title = "StretchDenoise: Parametric Curve Reconstruction with Guarantees by Separating Connectivity from Residual Uncertainty of Samples", author = "Stefan Ohrhallinger and Michael Wimmer", year = "2018", abstract = "We reconstruct a closed denoised curve from an unstructured and highly noisy 2D point cloud. Our proposed method uses a two-pass approach: Previously recovered manifold connectivity is used for ordering noisy samples along this manifold and express these as residuals in order to enable parametric denoising. This separates recovering low-frequency features from denoising high frequencies, which avoids over-smoothing. The noise probability density functions (PDFs) at samples are either taken from sensor noise models or from estimates of the connectivity recovered in the first pass. The output curve balances the signed distances (inside/outside) to the samples. Additionally, the angles between edges of the polygon representing the connectivity become minimized in the least-square sense. The movement of the polygon's vertices is restricted to their noise extent, i.e., a cut-off distance corresponding to a maximum variance of the PDFs. We approximate the resulting optimization model, which consists of higher-order functions, by a linear model with good correspondence. Our algorithm is parameter-free and operates fast on the local neighborhoods determined by the connectivity. %We augment a least-squares solver constrained by a linear system to also handle bounds. This enables us to guarantee stochastic error bounds for sampled curves corrupted by noise, e.g., silhouettes from sensed data, and we improve on the reconstruction error from ground truth. Source code is available online. An extended version is available at: https://arxiv.org/abs/1808.07778", month = aug, isbn = "978-3-03868-073-4", location = "Hong Kong", event = "Pacific Graphics 2018", editor = "H. Fu, A. Ghosh, and J. Kopf (Guest Editors)", booktitle = "Proceedings of Pacific Graphics 2018", pages = "1--4", keywords = "Denoising, Curve reconstruction, Optimization", URL = "https://www.cg.tuwien.ac.at/research/publications/2018/ohrhallinger_stefan-2018-pg/", } @bachelorsthesis{pointner_michael-2017-baa, title = "Multi-Focal Image Generation using Automatic Depth-Based Focus Selection", author = "Michael Pointner", year = "2017", abstract = "One of the most important objectives in photography is the sharpness of the whole image, which is not so easy to achieve because of the physical properties of a camera lens. Due to Google’s new Tango API, which enables the perception of depth information on smartphones, we have evaluated the usability of this new technology for the generation of all-sharp images through multi-focus with the Lenovo Phab 2 Pro as the first smartphone to support this technology.", month = sep, address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria", school = "Institute of Computer Graphics and Algorithms, Vienna University of Technology ", keywords = "all-in-focus, focus selection, image stitching", URL = "https://www.cg.tuwien.ac.at/research/publications/2017/pointner_michael-2017-baa/", } @bachelorsthesis{grossmann-2016-baa, title = "Extracting Sensor Specific Noise Models", author = "Nicolas Grossmann", year = "2017", abstract = "With the growing number of consumer-oriented depth sensors like the Kinect or the newly released Phab2Pro, the question of how precise these sensors are arises. In this thesis we want to evaluate the average noise in the generated depth measurements in both the axial direction and the lateral directions. As part of a two-part project this thesis will view the noise’s development with varying distance and angle. Finally, we will present and evaluate two models describing the noise behavior, with the first being derived from solely this thesis’ measurements and the second one being a combination of the previous model and a model of a colleague. This derived models can be used in a post-processing step to filter the generated depth images.", month = aug, note = "1", address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria", school = "Institute of Computer Graphics and Algorithms, Vienna University of Technology ", keywords = "noise model, surface reconstruction, sensor noise", URL = "https://www.cg.tuwien.ac.at/research/publications/2017/grossmann-2016-baa/", } @bachelorsthesis{koeppel-2016-baa, title = "Extracting Noise Models – considering X / Y and Z Noise", author = "Thomas K\"{o}ppel", year = "2017", abstract = "We have developed two different test setups allowing the characterization of noise in X, Y and Z direction for the KinectV2 and the Phab2Pro depth sensors. We have combined these two methods, generating a single noise model allowing a prediction of the amount of noise in specific areas of an image in the three respective directions at a certain distance and rotation. We have conducted two test setups and measured the noise from 900 mm to 3.100 mm for the generation of the noise models. The test setup of this thesis focused on determining the noise in X, Y and Z direction, covering the whole frustum of the respective depth sensor. In this thesis, Z noise was measured against a wall and X and Y noises were measured using a 3D chequerboard that was shifted through the room, allowing the above mentioned coverage of the whole frustum. Along the edges of the cells of the chequerboard, the X and Y noise was measured. The combined model was evaluated by using a solid cube to classify the quality of our noise model.", month = aug, note = "1", address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria", school = "Institute of Computer Graphics and Algorithms, Vienna University of Technology ", keywords = "noise model, surface reconstruction, sensor noise", URL = "https://www.cg.tuwien.ac.at/research/publications/2017/koeppel-2016-baa/", } @inproceedings{Radwan-2017-Occ, title = "Cut and Paint: Occlusion-Aware Subset Selection for Surface Processing", author = "Mohamed Radwan and Stefan Ohrhallinger and Elmar Eisemann and Michael Wimmer", year = "2017", abstract = "User-guided surface selection operations are straightforward for visible regions on a convex model. However, concave surfaces present a challenge because self-occlusions require multiple camera positions to get unobstructed views. Therefore, users often have to locate and switch to new unobstructed views in order to continue the operation. Our novel approach enables operations like painting or cutting in a single view, even on the backside of objects and for arbitrary depth complexity, with interactive performance. Continuous projection of a curve drawn in screen space onto the mesh guarantees seamless brush strokes or manifold cuts, unaffected by any occlusions. Our occlusion-aware surface-processing method enables a number of applications in an easy way. As examples, we show continuous painting on the surface, selecting regions for texturing, creating illustrative cutaways from nested models and animation of cutaways.", month = may, publisher = "Canadian Human-Computer Communications Society / Soci{\'e}t{\'e} canadienne du dialogue humain-machine", location = "Edmonton, Alberta, CA", event = "Graphics Interface 2017", doi = "10.20380/GI2017.11", booktitle = "Proceedings of Graphics Interface 2017", pages = "82--89", URL = "https://www.cg.tuwien.ac.at/research/publications/2017/Radwan-2017-Occ/", } @bachelorsthesis{reinwald-2017-baa, title = "Fast kNN in Screen Space on GPU", author = "Siegfried Reinwald", year = "2017", abstract = "A common problem in computer science is to construct a spatial data structure (octree, kd-tree) and use it to search for k-nearest neighbors (kNN). In surface reconstruction from dynamic points (real-time), both construction and search time are critical. As points on a surface are a sparse distribution in 3D, this can be exploited by mapping them into screen space (2D), as shown in "Auto-splats" (Preiner 2012). Our approach is to also exploit spatial coherence in screen space to find kNN for points. The performance is maximized by a CUDA implementation designed to minimize memory-boundness. An preliminary implementation exists which constructs a packed quadtree and reads kNN from a quadtree node density estimate. A few improvements have to be made to optimize performance and to analyze results.", month = mar, note = "1", address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria", school = "Institute of Computer Graphics and Algorithms, Vienna University of Technology ", keywords = "k-nearest neighbors, surface reconstruction, parallel optimization", URL = "https://www.cg.tuwien.ac.at/research/publications/2017/reinwald-2017-baa/", } @article{forsythe-2016-ccm, title = "Resolution-independent superpixels based on convex constrained meshes without small angles", author = "Jeremy Forsythe and Vitaliy Kurlin and Andrew Fitzgibbon", year = "2016", abstract = "The over-segmentation problem for images is studied in the new resolution-independent formulation when a large image is approximated by a small number of convex polygons with straight edges at subpixel precision. These polygonal superpixels are obtained by refining and extending subpixel edge segments to a full mesh of convex polygons without small angles and with approximation guarantees. Another novelty is the objective error difference between an original pixel-based image and the reconstructed image with a best constant color over each superpixel, which does not need human segmentations. The experiments on images from the Berkeley Segmentation Database show that new meshes are smaller and provide better approximations than the state-of-the-art.", month = dec, journal = "Lecture Notes in Computer Science (LNCS)", volume = "10072", issn = "0302-9743", pages = "223--233", keywords = "superpixels, polygonal mesh, Delaunay triangulation, constrained triangulation, edge detection", URL = "https://www.cg.tuwien.ac.at/research/publications/2016/forsythe-2016-ccm/", } @article{ohrhallinger-2016-sgp, title = "Curve Reconstruction with Many Fewer Samples", author = "Stefan Ohrhallinger and Scott A. Mitchell and Michael Wimmer", year = "2016", abstract = "We consider the problem of sampling points from a collection of smooth curves in the plane, such that the Crust family of proximity-based reconstruction algorithms can rebuild the curves. Reconstruction requires a dense sampling of local features, i.e., parts of the curve that are close in Euclidean distance but far apart geodesically. We show that epsilon<0.47-sampling is sufficient for our proposed HNN-CRUST variant, improving upon the state-of-the-art requirement of epsilon<1/3-sampling. Thus we may reconstruct curves with many fewer samples. We also present a new sampling scheme that reduces the required density even further than epsilon<0.47-sampling. We achieve this by better controlling the spacing between geodesically consecutive points. Our novel sampling condition is based on the reach, the minimum local feature size along intervals between samples. This is mathematically closer to the reconstruction density requirements, particularly near sharp-angled features. We prove lower and upper bounds on reach rho-sampling density in terms of lfs epsilon-sampling and demonstrate that we typically reduce the required number of samples for reconstruction by more than half. ", journal = "Computer Graphics Forum", volume = "35", number = "5", issn = "1467-8659", pages = "167--176", keywords = "sampling condition, curve reconstruction, curve sampling", URL = "https://www.cg.tuwien.ac.at/research/publications/2016/ohrhallinger-2016-sgp/", }