Rendering engine

This page provides some details on my personal academic rendering engine, started and improved over the course of my PhD, and continued for my current research and learning projects. The core source code is composed of 35000 lines of C++, it has been used to render thousands of hours of images and all figures from my academic publications, and used on various machines including the compute cluster IN2P3.
Unfortunately the codebase is not public yet due to two reasons: some unpublished work implemented in it and a lazyness in the code quality that makes some part of it obfuscated. Below are the list of features it includes and some renders produced by the renderer.
However, a small portion of it has been separated and published alongside my publication Volumetric multi-view rendering.

Features

It includes the following list of features:

Integrators
- Unidirectional volumetric path tracing integrator (vpt) [Miller:2019]
- Multiple view volumetric path tracing integrator (mvpt) and its adaptive version (amvpt) [Fraboni:2021]
- Multiple view path tracing integrator (extended to volumetric support for comparisons) [Fraboni:2019]
- Miscellaneous: Ambient occlusion, Normal, Light tracing, Path tracing, DoF supersampling integrator, Discrete path reusing integrator (surfaces only) [Bekaert:2002]
Materials
- smooth models: diffuse reflection/transmission, conductor, dielectric with absorption, plastic with absorption, bi-layered model (coating and substrate) with absorption
- rough models: diffuse reflection/transmission, conductor, dielectric with absorption, plastic with absorption, bi-layered model (coating and substrate) with absorption
- microfacet models Beckmann and GGX
- energy compensation [Turquin:2017][Kulla:2017]
- mix BSDF
- thin film interferences
- BSDF validation methods [Heitz:2014]
- BSDF similarity methods [Fraboni:2021]
- Roughness regularization [Kaplanyan:2013][Jendersie:2019]
Direct illumination techniques
- BSDF sampling
- Next event estimation [Shirley:1994]
- Equiangular sampling [Kulla:2012]
- Light tree importance sampling (diffuse + approximate specular) [Conty:2018] [Yuksel:2019] [Lin:2019]
- Multiple Importance Sampling [Veach:1995]
- RIS with CDF inversion or reservoirs (ReSTIR) [Talbot:2005][Bitterli:2020]
Homogeneous and heterogeneous participating media
- Spectral absorption, scattering and emission support
- Density functions
  - Uniform and 3D textures density
  - Sparse voxel density support with NanoVDB [Museth:2021]
  - Procedural 3D noise density (value, perlin, simplex, ridged) and 4D for animated media support with FastNoise [Peck:2020][Musgrave:1994]
- Phase functions
  - Uniform
  - Henyey-Greenstein
  - Mie approximation with numerical CDF inversion
- Transmittance estimators
  - Spectral delta tracking and spectral ratio tracking [Miller:2019]
  - Unbiased ray marching approximation [Kettunen:2021]
Textures and UV mapping
Depth of field with thin lens model [Kolb:1995]
Time integration for motion blur
SIMD math library
Fast ray - scene intersection with Intel Embree 4 [Wald:2014] (and Standalone BVH as well)
Progressive rendering with interactive OpenGL preview, faster offline batched tiles mode
Adaptive rendering with various error criteria [Fraboni:2021][Fraboni:2019][Kajiya:1986][Rigau:2003][Heitz:2018]:
- Variance error criterion [Kajiya:1986]
- f-divergence error criterion [Rigau:2003]
- Total variation noise estimation [Heitz:2018]

Examples


The dragon model courtesy of Stanford Computer Graphics Laboratory	The spaceship model courtesy of thecali
Lucy model courtesy of Stanford Computer Graphics Laboratory	The shogun model courtesy of Zeroswat
The Disney Cloud Data Set courtesy of Walt Disney Animation Studios	The sponza atrium model courtesy of Marko Dabrovic
Zero day model courtesy of Beeple	The pool table model courtesy of Chamouleau