VMV2021

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2633087

Browse

Browsing VMV2021 by Subject "Computing methodologies"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Capturing Anisotropic SVBRDFs
    (The Eurographics Association, 2021) Kaltheuner, Julian; Bode, Lukas; Klein, Reinhard; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    In this work, we adapt and improve recent isotropic material estimation efforts to estimate spatially varying anisotropic materials with an additional Fresnel term using a variable set of input images and are able to handle any resolution. We combine an initial estimation network with an auto-encoder to fine-tune the decoding of latent embedded appearance parameters on the input images to produce finely detailed SVBRDFs. For this purpose, the training must be adapted so that the determination is possible on the basis of a small number of images that still capture as much reflective behavior of materials as possible. The resulting appearance parameters are capable of capturing and reconstructing complex spatially varying features in detail, but place increased demands on the input images.
  • Thumbnail Image
    Item
    EMCA: Explorer of Monte Carlo based Algorithms
    (The Eurographics Association, 2021) Ruppert, Lukas; Kreisl, Christoph; Blank, Nils; Herholz, Sebastian; Lensch, Hendrik P. A.; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    Debugging or analyzing the performance of global illumination algorithms is a challenging task due to the complex path-scene interaction and numerous places where errors and programming bugs can occur. We present a novel, lightweight visualization tool to aid in the understanding of global illumination and the debugging of rendering frameworks. The tool provides detailed information about intersections and light transport paths. Users can add arbitrary data of their choosing to each intersection, based on their specific demands. Aggregate plots allow users to quickly discover and select outliers for further inspection across the globally linked visualization views. That information is further coupled with 3D visualization of the scene where additional aggregated information on the surfaces can be inspected in false colors. These include 3D heat maps such as the density of intersections as well as more advanced colorings such as a diffuse transport approximation computed from local irradiance samples and diffuse material approximations. The necessary data for the 3D coloring is collected as a side-product of quickly rendering the image at low sample counts without significantly slowing down the rendering process. It requires almost no precomputation and very little storage compared to point cloud-based approaches. We present several use cases of how novices and advanced rendering researchers can leverage the presented tool to speed up their research.
  • Thumbnail Image
    Item
    FERMIUM: A Framework for Real-time Procedural Point Cloud Animation and Morphing
    (The Eurographics Association, 2021) Wegen, Ole; Böttger, Florence; Döllner, Jürgen; Trapp, Matthias; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    This paper presents a framework for generating real-time procedural animations and morphing of 3D point clouds. Point clouds or point-based geometry of varying density can easily be acquired using LiDAR cameras or modern smartphones with LiDAR sensors. This raises the question how this raw data can directly be used in the creative industry to create novel digital content using animations. For this purpose, we describe a framework that enables the implementation and combination of animation effects for point clouds. It takes advantage of graphics hardware capabilities and enables the processing of complex datasets comprising up to millions of points. In addition, we compare and evaluate implementation variants for the subsequent morphing of multiple 3D point clouds.
  • Thumbnail Image
    Item
    Real-Time Curvature-aware Re-Parametrization and Tessellation of Bézier Surfaces
    (The Eurographics Association, 2021) Buchenau, Christoph; Guthe, Michael; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    Interactive tessellation of parametric surfaces has many applications in both engineering and entertainment computing. The most common primitives are bi-cubic Bézier patches which are, among others, an intermediate representation of subdivision surfaces for rendering. The current state-of-the-art employs hardware tessellation where a uniform subdivison pattern is used per patch. If the curvature varies strongly over a patch, this results in an over-tessellation of flat areas. Based on the observation that the second derivative changes linearly over the patch, we show that it is possible to reparameterize the patches such that the tessellation adapts to the curvature. This way, we reduce the number of primitives by an average of 15% for the same error bound.
  • Thumbnail Image
    Item
    SuBloNet: Sparse Super Block Networks for Large Scale Volumetric Fusion
    (The Eurographics Association, 2021) Rückert, Darius; Stamminger, Marc; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    Training and inference of convolutional neural networks (CNNs) on truncated signed distance fields (TSDFs) is a challenging task. Large parts of the scene are usually empty, which makes dense implementations inefficient in terms of memory consumption and compute throughput. However, due to the truncation distance, non-zero values are grouped around the surface creating small dense blocks inside the large empty space. We show that this structure can be exploited by storing the TSDF in a block sparse tensor and then decomposing it into rectilinear super blocks. A super block is a dense 3d cuboid of variable size and can be processed by conventional CNNs. We analyze the rectilinear decomposition and present a formulation for computing the bandwidth-optimal solution given a specific network architecture. However, this solution is NP-complete, therefore we also a present a heuristic approach for fast training and inference tasks. We verify the effectiveness of SuBloNet and report a speedup of 4x towards dense implementations and 1.7x towards state-of-the-art sparse implementations. Using the super block architecture, we show that recurrent volumetric fusion is now possible on large scale scenes. Such a systems is able to reconstruct high-quality surfaces from few noisy depth images.