Physically Based Rendering

Overview

A set of physically based rendering experiments built on the pbrt-v4 reference renderer during university labs. Instead of touching the renderer’s core, I authored and configured scenes to study how modern light transport behaves: comparing integrators, materials and participating media, then analysing how each parameter moved noise, image quality and render time.

Technical Highlights

• Material and light-transport study. A Cornell-box-style scene instances one mesh nine times with different BxDFs (diffuse, dielectric, textured) under identical lighting, isolating how each material responds. See scene.pbrt.
• Volumetric media. A participating-media scene lit by a spotlight, rendered with pbrt’s volumetric path integrator to study fog and scattering. Output in spotFog.png.
• Path tracing versus bidirectional. The same setup rendered with unidirectional and bidirectional path tracing to compare noise and convergence. See bidir.png.
• Spectral rendering. Scenes authored for pbrt’s sampled-spectrum pipeline instead of RGB, combining procedural and image-based textures. See template/scene.pbrt.
• GPU rendering with CUDA and OptiX. The build was configured for GPU-accelerated rendering, with a live tev display server and progressive samples-per-pixel for fast iteration. See commands.txt.

Learnings

The value here was less about writing renderer code and more about reading how a production-grade renderer is put together, and learning to reason about rendering choices from measured output rather than guesswork.