VMV15

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/13268

Browse

Now showing 1 - 20 of 22

Variational Separation of Light Field Layers
(The Eurographics Association, 2015) Johannsen, Ole; Sulc, Antonin; Goldluecke, Bastian; David Bommes and Tobias Ritschel and Thomas Schultz
Images of scenes which contain reflective or transparent surfaces are composed of different layers which are observed at different depths. Analyzing such a scene requires separating the image into its individual layers, which remains a challenging and important problem. While the problem is very much ill-posed when only a single image is considered, recent work has shown that depth estimation for two layers becomes quite tractable when one instead captures a 4D light field of the scene. In this paper, we propose a novel variational approach to layer separation which is based on these ideas. We formulate a linear generative model to reconstruct the light field from disparity and luminance information for the individual layers on the center view. Comparing the model with the observerd data yields a convex variational problem for layer reconstruction, which can be solved to global optimality with a primal-dual scheme. Layer disparity is estimated in a first step, for which we improve upon a model based on second order structure tensors on the epipolar plane images. In contrast to previous work, the resulting approach is robust enough to be able to deal with light fields from the Lytro Illum camera, for which we obtain a compelling separation of the reflectance layer in real-world scenes.
Frontmatter: Vision, Modeling and Visualization (VMV)
(The Eurographics Association, 2015) Bommes, David; Ritschel, Tobias; Schultz, Thomas; -
Point-wise Map Recovery and Refinement from Functional Correspondence
(The Eurographics Association, 2015) Rodolà, Emanuele; Moeller, Michael; Cremers, Daniel; David Bommes and Tobias Ritschel and Thomas Schultz
Since their introduction in the shape analysis community, functional maps have met with considerable success due to their ability to compactly represent dense correspondences between deformable shapes. Despite the numerous advantages of such representation, however, the problem of converting a given functional map back to a point-topoint map has received a surprisingly limited interest. In this paper we analyze the general problem of point-wise map recovery from arbitrary functional maps. In doing so, we rule out many of the assumptions required by the currently established approach - most notably, the limiting requirement of the input shapes being nearlyisometric. We devise an efficient recovery process based on a simple probabilistic model. Experiments confirm that this approach achieves remarkable accuracy improvements in very challenging cases.
The Bounced Z-buffer for Indirect Visibility
(The Eurographics Association, 2015) Nalbach, Oliver; Ritschel, Tobias; Seidel, Hans-Peter; David Bommes and Tobias Ritschel and Thomas Schultz
Synthesizing images of animated scenes with indirect illumination and glossy materials at interactive frame rates commonly ignores indirect shadows. In this work we extend a class of indirect lighting algorithms that splat shading to a framebuffer - we demonstrate deep screen space and ambient occlusion volumes - to include indirect visibility. To this end we propose the bounced z-buffer: While a common z-buffered framebuffer, at each pixel, maintains the distance from the closest surface and its radiance along a direction from the camera to that pixel, our new representation contains the distance from the closest surface and its radiance after one indirect bounce into a certain other direction. Consequently, with bounced z-buffering, only the splat from the nearest emitter in one direction contributes to each pixel. Importance-sampling the bounced directions according to the product of cosine term and BRDF allows to approximate full shading by a simple sum of neighboring framebuffer pixels.
Vector-to-Closest-Point Octree for Surface Ray-Casting
(The Eurographics Association, 2015) Demir, Ismail; Westermann, Rüdiger; David Bommes and Tobias Ritschel and Thomas Schultz
GPU voxel-based surface ray-casting has positioned as an interesting alternative to rasterization-based rendering approaches, because it allows using many processing units simultaneously, can effectively exploit thread level parallelism, and enables fine-granularity occlusion culling on the pixel level. Yet voxel-based techniques face the problem that an extremely high resolution is necessary to avoid block artifacts at high zoom levels. In this work, we propose a novel improvement of voxel-based ray-casting to overcome this limitation. By using a hierarchical Vector-to-Closest-Point (VCP) representation, we can inherit the advantages of a voxel-based approach at a much smoother approximation of the surface. We demonstrate that, although the VCP grid consumes more memory per cell, it requires less memory overall, because it builds upon a significantly shallower tree hierarchy. In a number of examples we demonstrate the use of our approach for high-quality rendering of high resolution surface models.
A Convex Clustering-based Regularizer for Image Segmentation
(The Eurographics Association, 2015) Hell, Benjamin; Magnor, Marcus; David Bommes and Tobias Ritschel and Thomas Schultz
In this paper we present a novel way of combining the process of k-means clustering with image segmentation by introducing a convex regularizer for segmentation-based optimization problems. Instead of separating the clustering process from the core image segmentation algorithm, this regularizer allows the direct incorporation of clustering information in many segmentation algorithms. Besides introducing the model of the regularizer, we present a numerical algorithm to efficiently solve the occurring optimization problem while maintaining complete compatibility with any other gradient descent based optimization method. As a side-product, this algorithm also introduces a new way to solve the rather elaborate relaxed k-means clustering problem, which has been established as a convex alternative to the non-convex k-means problem.
Efficient GPU Based Sampling for Scene-Space Video Processing
(The Eurographics Association, 2015) Klose, Felix; Wang, Oliver; Bazin, Jean-Charles; Magnor, Marcus; Sorkine-Hornung, Alexander; David Bommes and Tobias Ritschel and Thomas Schultz
We describe a method to efficiently collect and filter a large set of 2D pixel observations of unstructured 3D points, with applications to scene-space aware video processing. One of the main challenges in scene-space video processing is to achieve reasonable computation time despite the very large volumes of data, often in the order of billions of pixels. The bottleneck is determining a suitable set of candidate samples used to compute each output video pixel color. These samples are observations of the same 3D point, and must be gathered from a large number of candidate pixels, by volumetric 3D queries in scene-space. Our approach takes advantage of the spatial and temporal continuity inherent to video to greatly reduce the candidate set of samples by solving 3D volumetric queries directly on a series of 2D projections, using out-of-core data streaming and an efficient GPU producerconsumer scheme that maximizes hardware utilization by exploiting memory locality. Our system is capable of processing over a trillion pixel samples, enabling various scene-space video processing applications on full HD video output with hundreds of frames and processing times in the order of a few minutes.
Fast Multiplexed Acquisition of High-dynamic-range Material Appearance
(The Eurographics Association, 2015) Brok, Dennis den; Steinhausen, Heinz Christian; Klein, Reinhard; David Bommes and Tobias Ritschel and Thomas Schultz
There is tremendous demand for digital representations that allow for materials to be re-lit under as wide a range of illumination scenarios as possible. It is therefore desirable to capture the entire dynamic range of a material's appearance. This process can require excessive shutter times for many materials that reflect only small amounts of light for certain lighting and viewing directions, for instance in the presence of low albedo or self-shadowing. The problem is amplified in the case of image-based appearance models such as the bidirectional texture function (BTF), where possibly many thousands of images are required to accurately sample high-frequency details in the angular domain. We propose to capture material BTFs with their dynamic range compressed by multiplexed illumination. We demonstrate that the signal-dependent noise associated with demultiplexing can be mitigated by means of an existing database of low-noise material BTFs. Moreover, we investigate a method to quickly create a suitable database from scratch.
Level-of-Detail for Production-Scale Path Tracing
(The Eurographics Association, 2015) Prus, Magdalena; Eisenacher, Christian; Stamminger, Marc; David Bommes and Tobias Ritschel and Thomas Schultz
Path-traced global illumination (GI) becomes increasingly important in movie production. With offscreen elements considerably contributing to the path traced image, geometric complexity increases drastically, requiring geometric instancing or a variety of manually created and baked LOD. To reduce artists' work load and bridge the gap between mesh-based LOD (Mip-maps) and voxel-based LOD (brickmaps), we propose to use an SVO with averaged BRDF parameters, e.g. for the Disney-BRDF, and a normal distribution per voxel. During shading we construct a BRDF from the averaged BRDF parameters and evaluate it with a random normal sampled from the distribution. This is simple, memory-efficient, and handles a wide variety of geometry scales and materials seamlessly, with proper filtering. Further it is efficient to construct, which allows quick artist iterations as well as automatic and lazy generation on scene loading.
A Taxonomy of Integration Techniques for Spatial and Non-Spatial Visualizations
(The Eurographics Association, 2015) Sorger, Johannes; Ortner, Thomas; Piringer, Harald; Hesina, Gerd; Gröller, Eduard; David Bommes and Tobias Ritschel and Thomas Schultz
Research on visual data representations is traditionally classified into methods assuming an inherent mapping from data values to spatial coordinates (scientific visualization and real-time rendering) and methods for abstract data lacking explicit spatial references (information visualization). In practice, however, many applications need to analyze data comprising abstract and spatial information, thereby spanning both visualization domains. Traditional classification schemes do not support a formal description of these integrated systems. The contribution of this paper is a taxonomy that describes a holistic design space for integrating components of spatial and abstract visualizations. We structure a visualization into three components: Data, Visual, and Navigation. These components can be linked to build integrated visualizations. Our taxonomy provides an alternative view on the field of visualization in a time where the border between scientific and information visualization becomes blurred.
Rotoscoping on Stereoscopic Images and Videos
(The Eurographics Association, 2015) Bukenberger, Dennis R.; Schwarz, Katharina; Groh, Fabian; Lensch, Hendrik P. A.; David Bommes and Tobias Ritschel and Thomas Schultz
Creating an animation based on video footage (rotoscoping) often requires significant manual work. For monoscopic videos diverse publications already feature (semi-)automatic techniques to apply non-photorealistic image abstraction (NPR) to videos. This paper addresses abstraction of 3D stereo content minimizing stereoscopic discomfort in images and videos. We introduce a completely autonomous framework that enhances stereo and temporal consistency. Stereoscopic coherence with consistent textures for both eyes is produced by warping the left and right images into a central disparity domain followed by mapping them back to the left and right view. Smooth movements with reduced flickering are achieved by considering optical flow in the propagation of abstract features between frames. The results show significant improvements of stereo consistency without discomforting artifacts in the depth perception. We extend existing stroke based rendering (SBR) for higher accuracy at strong image gradients. Furthermore, we demonstrate that our stereo framework is easily applicable to other point-based abstraction styles. Finally, we evaluate the stereo consistency of our results in a small user study and show that the comfort of the visual appearance is maintained.
Data Driven 3D Face Tracking Based on a Facial Deformation Model
(The Eurographics Association, 2015) Sibbing, Dominik; Kobbelt, Leif; David Bommes and Tobias Ritschel and Thomas Schultz
We introduce a new markerless 3D face tracking approach for 2D video streams captured by a single consumer grade camera. Our approach is based on tracking 2D features in the video and matching them with the projection of the corresponding feature points of a deformable 3D model. By this we estimate the initial shape and pose of the face. To make the tracking and reconstruction more robust we add a smoothness prior for pose changes as well as for deformations of the faces. Our major contribution lies in the formulation of the smooth deformation prior which we derive from a large database of previously captured facial animations showing different (dynamic) facial expressions of a fairly large number of subjects. We split these animation sequences into snippets of fixed length which we use to predict the facial motion based on previous frames. In order to keep the deformation model compact and independent from the individual physiognomy, we represent it by deformation gradients (instead of vertex positions) and apply a principal component analysis in deformation gradient space to extract the major modes of facial deformation. Since the facial deformation is optimized during tracking, it is particularly easy to apply them to other physiognomies and thereby re-target the facial expressions. We demonstrate the effectiveness of our technique on a number of examples.
Hierarchical Hashing for Pattern Search in 3D Vector Fields
(The Eurographics Association, 2015) Wang, Zhongjie; Seidel, Hans-Peter; Weinkauf, Tino; David Bommes and Tobias Ritschel and Thomas Schultz
The expressiveness of many visualization methods for 3D vector fields is often limited by occlusion, i.e., interesting flow patterns hide each other or are hidden by laminar flow. Automatic detection of patterns in 3D vector fields has gained attention recently, since it allows to highlight user-defined patterns and separate the wheat from the chaff. We propose an algorithm which is able to detect 3D flow patterns of arbitrary extent in a robust manner. We encode the local flow behavior in scale space using a sequence of hierarchical base descriptors, which are pre-computed and hashed into a number of hash tables. This ensures a fast fetching of similar occurrences in the flow and requires only a constant number of table lookups. In contrast to many previous approaches, our method supports patterns of arbitrary shape and extent. We achieve this by assembling these patterns using several smaller spheres. The results are independent of translation, rotation, and scaling. Our experiments show that our approach encompasses the state of the art with respect to both the computational costs and the accuracy.
Accurate Face Reconstruction through Anisotropic Fitting and Eye Correction
(The Eurographics Association, 2015) Achenbach, Jascha; Zell, Eduard; Botsch, Mario; David Bommes and Tobias Ritschel and Thomas Schultz
Fitting a facial template model to 3D-scanner data is a powerful technique for generating face avatars, in particular in the presence of noisy and incomplete measurements. Consequently, there are many approaches for the underlying non-rigid registration task, and these are typically composed from very similar algorithmic building blocks. By providing a thorough analysis of the different design choices, we derive a face matching technique tailored to high quality reconstructions from high resolution scanner data. We then extend this approach in two ways: An anisotropic bending model allows us to more accurately reconstruct facial details. A simultaneous constrained fitting of eyes and eye lids improves the reconstruction of the eye region considerably.
Temporally Consistent Wide Baseline Facial Performance Capture via Image Warping
(The Eurographics Association, 2015) Kettern, Markus; Hilsmann, Anna; Eisert, Peter; David Bommes and Tobias Ritschel and Thomas Schultz
In this paper, we present a method for detailed temporally consistent facial performance capture that supports any number of arbitrarily placed video cameras. Using a suitable 3D model as reference geometry, our method tracks facial movement and deformation as well as photometric changes due to illumination and shadows. In an analysis-by-synthesis framework, we warp one single reference image per camera to all frames of the sequence thereby drastically reducing temporal drift which is a serious problem for many state-of-the-art approaches. Temporal appearance variations are handled by a photometric estimation component modeling local intensity changes between the reference image and each individual frame. All parameters of the problem are estimated jointly so that we do not require separate estimation steps that might interfere with one another.
Temporal Coherence Predictor for Time Varying Volume Data Based on Perceptual Functions
(The Eurographics Association, 2015) Noonan, Tom; Campoalegre, Lazaro; Dingliana, John; David Bommes and Tobias Ritschel and Thomas Schultz
This paper introduces an empirical, perceptually-based method which exploits the temporal coherence in consecutive frames to reduce the CPU-GPU traffic size during real-time visualization of time-varying volume data. In this new scheme, a multi-threaded CPU mechanism simulates GPU pre-rendering functions to characterize the local behaviour of the volume. These functions exploit the temporal coherence in the data to reduce the sending of complete per frame datasets to the GPU. These predictive computations are designed to be simple enough to be run in parallel on the CPU while improving the general performance of GPU rendering. Tests performed provide evidence that we are able to reduce considerably the texture size transferred at each frame without losing visual quality while maintaining performance compared to the sending of entire frames to the GPU. The proposed framework is designed to be scalable to Client/Server network based implementations to deal with multi-user systems.
Tongue S(t)imulator - A Comprehensive Parametrized Pose Model for Speech Therapy
(The Eurographics Association, 2015) Haraké, Laura; Bełtkiewicz, Dorota; Lochmann, Gerrit; David Bommes and Tobias Ritschel and Thomas Schultz
Recent digital applications in speech therapy address patients to train auditive speech comprehension, reading or semantics, in a playful way. Virtual tutors consist of three-dimensional head models for assisting the patient with conversational exercises. However, speech therapists also have to give pronunciation instructions and motility training of the tongue very often, but only have two-dimensional drawings or their own mouths for demonstration. In this paper we propose a comprehensive application for speech therapy as a therapeutical tool, simulating the articulation of German phones including color-coded expiration flows and the deglutition process (swallowing). A three-dimensional visualization of anatomical models of pharyngo-laryngeal area can be used in an interactive way. For examining the benefits of our tool over common conventional therapy media, our approach considers iteratively the demands of speech therapists. A final expert interview was conducted to assess how the application could be involved in treatment and the application’'s limits.
Interactive GPU-based Visualization of Scalar Data with Gaussian Distributed Uncertainty
(The Eurographics Association, 2015) Schlegel, Steven; Goldau, Mathias; Scheuermann, Gerik; David Bommes and Tobias Ritschel and Thomas Schultz
We present a GPU-based approach to visualize samples of normally distributed uncertain, three-dimensional scalar data. Our approach uses a mathematically sound interpolation scheme, i.e., Gaussian process regression. The focus of this work is to demonstrate, that GP-regression can be used for interpolation in practice, despite the high computational costs. The potential of our method is demonstrated by an interactive volume rendering of three-dimensional data, where the gradient estimation is directly computed by the field function without the need of additional sample points of the underlying data. We illustrate our method using three-dimensional data sets of the medical research domain.
Simple, Robust, Constant-Time Bounds on Surface Geodesic Distances using Point Landmarks
(The Eurographics Association, 2015) Burghard, Oliver; Klein, Reinhard; David Bommes and Tobias Ritschel and Thomas Schultz
In this paper we exploit redundant information in geodesic distance fields for a quick approximation of all-pair distances. Starting with geodesic distance fields of equally distributed landmarks we analyze the lower and upper bound resulting from the triangle inequality and show that both bounds converge reasonably fast to the original distance field. The lower bound has itself a bounded relative error, fulfills the triangle equation and under mild conditions is a distance metric. While the absolute error of both bounds is smaller than the maximal landmark distances, the upper bound often exhibits smaller error close to the cut locus. Both the lower and upper bound are simple to implement and quickly to evaluate with a constant-time effort for point-to-point distances, which are often required by various algorithms.
Extrapolating Large-Scale Material BTFs under Cross-Device Constraints
(The Eurographics Association, 2015) Steinhausen, Heinz Christian; Brok, Dennis den; Hullin, Matthias B.; Klein, Reinhard; David Bommes and Tobias Ritschel and Thomas Schultz
In this paper, we address the problem of acquiring bidirectional texture functions (BTFs) of large-scale material samples. Our approach fuses gonioreflectometric measurements of small samples with few constraint images taken on a flatbed scanner under semi-controlled conditions. Underlying our method is a lightweight texture synthesis scheme using a local texture descriptor that combines shading and albedo across devices. Since it operates directly on SVD-compressed BTF data, our method is computationally efficient and can be implemented on a moderate memory footprint.

Browse

Browsing VMV15 by Issue Date

Results Per Page

Sort Options