Alen Kurtagić
Supervisor(s): Ciril Bohak, Žiga Lesar
University of Ljubljana
Abstract: Geospatial point-based datasets commonly exceed billions of points. Real-time rendering therefore requires out-of core streaming, level of detail techniques, and numerically stable handling of Earth-scale coordinates. This paper presents a cross-platform C++ point-based rendering en gine built on WebGPU. The engine follows a data-oriented design and supports streaming of 3D Tiles datasets. To maintain numerical stability at geospatial scales, the sys tem uses camera-relative rendering using double-precision on CPU. Point cloud rasterization follows a multi-pass ren dering pipeline coordinated by a render graph. To reduce the overhead of frequent GPU memory allocation during streaming, the engine includes a buffer suballocator that reduces allocation overhead compared to direct WebGPU buffer creation. The resulting system enables real-time rendering of massive point clouds on consumer hardware.
Keywords: Point-based Graphics, Real-time Graphics, Rendering
Full text: Year: 2026
