Zusammenfassung:
This paper is about visualizing large matrices, also called datacubes,
in real time (RT) using Web technologies and the power of either
CPU-based or dedicated graphic processor unit (GPU). The aim is to
propose a convenient palette of volume-rendering models for the scientific
community and especially data scientists. Therefore, we have oriented our
approach to a Python Toolbox and Jupyter Notebook deployment. These
models are developed in JavaScript using the WebGL 1.0 graphics API.
We realized the rendering of these datacubes thanks to an improved
raycasting algorithm that allows dimensions of 16^3, 64^3, 256^3, and up to
1024^3, corresponding to a billion cells. The toolbox allows three rendering
types: (1) X-Ray for 3D datacubes in a (x, y, z) space; (2) a model simulating
the implicit surfaces for 3D datacubes in a (x, y, z) space; (3) and
last, an original model, we named Derived-rendering, for 3D datacubes
in a (x, y, t) space where t represents time. We also introduce solutions
to reduce the memory footprint and load on the GPU side. Tested with
a nowadays hardware configuration, our proposal demonstrates we can
even reach RT rendering for a billion cells datacube.
Keywords: Massive Data · 3D visualization · Datacube · Raycasting ·
Real-time · WebGL · GPU · GPGPU · Jupiter notebook · Python.