High-Performance Graphics 2021 - Symposium Papers

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2633077

Browse

Browsing High-Performance Graphics 2021 - Symposium Papers by Subject "Computing methodologies"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Compression and Rendering of Textured Point Clouds via Sparse Coding
    (The Eurographics Association, 2021) Schuster, Kersten; Trettner, Philip; Schmitz, Patric; Schakib, Julian; Kobbelt, Leif; Binder, Nikolaus and Ritschel, Tobias
    Splat-based rendering techniques produce highly realistic renderings from 3D scan data without prior mesh generation. Mapping high-resolution photographs to the splat primitives enables detailed reproduction of surface appearance. However, in many cases these massive datasets do not fit into GPU memory. In this paper, we present a compression and rendering method that is designed for large textured point cloud datasets. Our goal is to achieve compression ratios that outperform generic texture compression algorithms, while still retaining the ability to efficiently render without prior decompression. To achieve this, we resample the input textures by projecting them onto the splats and create a fixed-size representation that can be approximated by a sparse dictionary coding scheme. Each splat has a variable number of codeword indices and associated weights, which define the final texture as a linear combination during rendering. For further reduction of the memory footprint, we compress geometric attributes by careful clustering and quantization of local neighborhoods. Our approach reduces the memory requirements of textured point clouds by one order of magnitude, while retaining the possibility to efficiently render the compressed data.
  • Thumbnail Image
    Item
    Multi-Resolution Shared Representative Filtering for Real-Time Depth Completion
    (The Eurographics Association, 2021) Wu, Yu-Ting; Li, Tzu-Mao; Shen, I-Chao; Lin, Hong-Shiang; Chuang, Yung-Yu; Binder, Nikolaus and Ritschel, Tobias
    We present shared representative filtering for real-time high-resolution depth completion with RGB-D sensors. Conventional filtering-based methods face a dilemma when the missing regions of the depth map are large. When the filter window is small, the filter fails to include enough samples. On the other hand, when the window is large, the method could oversmooth depth boundaries due to the error introduced by the extra samples. Our method adapts the filter kernels to the shape of the missing regions to collect a sufficient number of samples while avoiding oversmoothing. We collect depth samples by searching for a small set of similar pixels, which we call the representatives, using an efficient line search algorithm. We then combine the representatives using a joint bilateral weight. Experiments show that our method can filter a high-resolution depth map within a few milliseconds while outperforming previous filtering-based methods on both real-world and synthetic data in terms of both efficiency and accuracy, especially when dealing with large missing regions in depth maps.
  • Thumbnail Image
    Item
    Rearchitecting Spatiotemporal Resampling for Production
    (The Eurographics Association, 2021) Wyman, Chris; Panteleev, Alexey; Binder, Nikolaus and Ritschel, Tobias
    Recent work by Bitterli et al. [BWP*20] introduced a real-time, many-light algorithm for rendering dynamic direct illumination from millions of lights by iteratively applying resampled importance sampling using weighted reservoir sampling. While enabling new levels of lighting complexity in real-time, the total cost remained beyond the budgets of even the most computationally demanding games. We introduce key algorithmic improvements developed while productizing this method that collectively reduce lighting costs by up to 7x, dramatically improve memory coherence, shrink the required ray budget, increase rendering quality, and expose parameters that enable trading quality for performance.
  • Thumbnail Image
    Item
    Transfer-Function-Independent Acceleration Structure for Volume Rendering in Virtual Reality
    (The Eurographics Association, 2021) Faludi, Balázs; Zentai, Norbert; Zelechowski, Marek; Zam, Azhar; Rauter, Georg; Griessen, Mathias; Cattin, Philippe C.; Binder, Nikolaus and Ritschel, Tobias
    Visualizing volumetric medical datasets in a virtual reality environment enhances the sense of scale and has a wide range of applications in diagnostics, simulation, training, and surgical planning. To avoid motion sickness, rendering at the native refresh rate of the head-mounted display is important, and frame drops have to be avoided. Despite these strict requirements and the high computational complexity of direct volume rendering, it is feasible to provide a comfortable experience using volume ray casting on modern hardware. Many implementations use precomputed gradients or illumination to achieve the targeted frame rate, and most rely on acceleration structures, such as distance maps or octrees, to speed up the ray marching shader. With many of these techniques, the opacity of voxels is baked into the precomputed data, requiring a recomputation when the opacity changes. This makes it difficult to implement features that lead to a sudden change in voxel opacity, such as real-time transfer function editing, transparency masking, or toggling the visibility of segmented tissues. In this work, we present an empty space skipping technique using an octree that does not have to be recomputed when the transfer function is changed and performs well even when more complex transfer functions are used. We encode the content of the volume as bitfields in the octree and are able to skip empty areas, even with transfer functions that cannot efficiently be represented as a simple range of voxel values. We show that our approach allows arbitrarily editing of the transfer function in real-time while maintaining the target frame rate of 90 Hz.
  • Thumbnail Image
    Item
    Vertex-Blend Attribute Compression
    (The Eurographics Association, 2021) Kuth, Bastian; Meyer, Quirin; Binder, Nikolaus and Ritschel, Tobias
    Skeleton-based animations require per-vertex attributes called vertex-blend attributes. They consist of a weight tuple and a bone index tuple. With meshes becoming more complex, vertex-blend attributes call for compression. However, no technique exists that exploits their special properties. To this end, we propose a novel and optimal weight compression method called Optimal Simplex Sampling and a novel bone index compression. For our test models, we compress bone index tuples between 2.3:1 and 3.5:1 and weight tuples between 1.6:1 and 2.5:1 while being visually lossless. We show that our representations can speed rendering and reduces GPU memory requirements over uncompressed representations with a similar error. Further, our representations compress well with general-purpose codecs making them suitable for offline-storage and streaming.