ZHW: A Numerical CODEC for Big Data Scientific Computation
Explore the source record for details and available documents.
Engineering topics
Publications and source records attributed to Lindstrom, P.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Abstract not provided
Explore the source record for details and available documents.
Abstract not provided
Abstract not provided
Abstract not provided
Abstract not provided
Abstract not provided
We propose LR (Laced Ring) - a simple data structure for representing the connectivity of manifold triangle meshes. LR provides the option to store on average either 1.08 references per triangle or 26.2 bits per triangle. Its construction, from an input mesh that supports constant-time adjacency queries, has linear space and time complexity, and involves ordering most vertices along a nearly-Hamiltonian cycle. LR is best suited for applications that process meshes with fixed connectivity, as any changes to the connectivity require the data structure to be rebuilt. We provide an implementation of the set of standard random-access, constant-time operators for traversing a mesh, and show that LR often saves both space and traversal time over competing representations.
We describe a simple and efficient algorithm for two-view triangulation of 3D points from approximate 2D matches based on minimizing the L2 reprojection error. Our iterative algorithm improves on the one by Kanatani et al. by ensuring that in each iteration the epipolar constraint is satisfied. In the case where the two cameras are pointed in the same direction, the method provably converges to an optimal solution in exactly two iterations. For more general camera poses, two iterations are sufficient to achieve convergence to machine precision, which we exploit to devise a fast, non-iterative method. The resulting algorithm amounts to little more than solving a quadratic equation, and involves a fixed, small number of simple matrixvector operations and no conditional branches. We demonstrate that the method computes solutions that agree to very high precision with those of Hartley and Sturm's original polynomial method, though achieves higher numerical stability and 1-4 orders of magnitude greater speed.
We present a streaming geometry compression codec for multiresolution, uniformly-gridded, triangular terrain patches that supports very fast decompression. Our method is based on linear prediction and residual coding for lossless compression of the full-resolution data. As simplified patches on coarser levels in the hierarchy already incur some data loss, we optionally allow further quantization for more lossy compression. The quantization levels are adaptive on a per-patch basis, while still permitting seamless, adaptive tessellations of the terrain. Our geometry compression on such a hierarchy achieves compression ratios of 3:1 to 12:1. Our scheme is not only suitable for fast decompression on the CPU, but also for parallel decoding on the GPU with peak throughput over 2 billion triangles per second. Each terrain patch is independently decompressed on the fly from a variable-rate bitstream by a GPU geometry program with no branches or conditionals. Thus we can store the geometry compressed on the GPU, reducing storage and bandwidth requirements throughout the system. In our rendering approach, only compressed bitstreams and the decoded height values in the view-dependent 'cut' are explicitly stored on the GPU. Normal vectors are computed in a streaming fashion, and remaining geometry and texture coordinates, as well as mesh connectivity, are shared and re-used for all patches. We demonstrate and evaluate our algorithms on a small prototype system in which all compressed geometry fits in the GPU memory and decompression occurs on the fly every rendering frame without any cache maintenance.
We address the problem of estimating the relative pose, i.e. translation and rotation, of two calibrated cameras from image point correspondences. Our approach is to factor the nonlinear algebraic pose error functional into translational and rotational components, and to optimize translation and rotation independently. This factorization admits subproblems that can be solved using direct methods with practical guarantees on global optimality. That is, for a given translation, the corresponding optimal rotation can directly be determined, and vice versa. We show that these subproblems are equivalent to computing the least eigenvector of second- and fourth-order symmetric tensors. When neither translation or rotation is known, alternating translation and rotation optimization leads to a simple, efficient, and robust algorithm for pose estimation that improves on the well-known 5- and 8-point methods.