Spatial rendering cost profiling and visualization framework using spherical harmonics cost probes

This dataset contains the software developed for the master's thesis "Visualizations for Guiding Rendering Costs" (Danilo José Bettencourt de Freitas, Delft University of Technology, 2026).

Objective.

Stable frame rates are critical to real-time rendering, but conventional frame profilers only reveal when and why a frame exceeds its budget — not where in the scene, or along which view direction, the bottleneck lies. Developers relying on manual scene exploration or fixed camera paths risk missing expensive viewpoints or entire regions. This software implements a probe-based framework that captures view-dependent rendering cost across a static 3D scene and surfaces the expensive viewpoints through coordinated visualizations.

Type of research.

Applied computer-graphics / tool-building research, evaluated qualitatively through a working implementation and a set of rendering case studies rather than a controlled user study.

Methods.

A uniform grid of probes samples rendering cost over view directions on a sphere and encodes each probe's signal as memory-efficient Spherical Harmonics (SH) coefficients; a separate confirmation pass recovers each probe's true peak cost free of SH band-limiting error. Six linked visual idioms then guide the user from a global cost overview down to the specific expensive viewpoint. The framework is integrated into a custom Vulkan renderer and evaluated on case studies covering common bottlenecks — expensive materials, over-submitted draw calls from misconfigured occlusion culling, and vertex bottlenecks from missing level-of-detail.

Contents.

The complete C++ source of the Vulkan rendering engine and the SH cost-probe profiling/visualization framework; the primary thesis application; and four built-in procedural case-study scenes (overdraw, geometry, dungeon, peaks) that reproduce the evaluated bottlenecks. Also included: GLSL shader sources, glTF/GLB example assets and .scene configuration files, unit tests, and the full pinned build toolchain (Nix flake + CMake + Makefile), plus a README with build/run instructions, an MIT license, and citation metadata. Building and running require a Linux system (X11 or Wayland with XWayland) and a Vulkan-capable GPU.