WurstRenderer is a CPU-based rendering system I made in early 2019. It focuses on experimentability and readability. For instance, the length of the bidirectional path tracer code is only 430 lines long (included comments, debug code, a lot of blank lines). All integrators (3d) are compared against reference generated from Mitsuba. The result of convergence plots indicates that the implementation of all integrators is correct.
To avoid accidentally releasing code written by collaborators or sensitive code, all changes after mid 2019 were removed.
In this framework, it supports:
- Real-time viewer
- Obj importing and partially PBRT importing.
- Oren-Nayar BSDF
- Mixed BRDF
- Modified Phong [Lafortune and Willems 1994]
- Blue Dithering Mask Generation via Simulated Annealing [Georgiev and Fajardo 2016]
- Bidirectional Path Tracing [Veach's thesis 1997]
- Matrix Bidirectional Path Tracing [Chaitanya et al. 2018] (haven't verified the correctness yet)
- Path Tracing Next Event Estimation with MIS [Veach's thesis 1997]
- Direct RayTracing of Phong Tessellation [Ogaki and Tokuyoshi 2011]
- Precomputed Radiance Transfer (Diffuse) [Sloan et al. 2002]
- Precomputed Radiance Transfer (Glossy) [Sloan et al. 2002]
- Primary Sample Space Metropolis Light Transport [Kelemen and Szirmay-Kalos 2001]
- Environment Map with several representation includes equirectangular, spherical harmonics, spherical gaussians, and spherical fibonacci (thanks to inverse mapping [Keinert et al. 2015]).
- Spherical Gaussian Lobes Fitting via LibTorch
- Homogeneous Volumetric Rendering with Isotropic & Henyey-Greenstein phase functions.
Note: It seems that homogeneous volumetric rendering is broken again...
Flatland is the world where two-dimensional creatures live in. It is quite useful for understanding complicated algorithms. The renderer supports two integrators:
- Flatland Path Tracing
- Flatland PSSMLT
It simply takes the 3d scene and automatically take a slice along Z-Axis to create a 2d scene.
For flatland integrator, it can preview path density as well. I found it is very helpful for understanding MCMC & path guiding experiments.
 |
 |
 |
 |
|---|---|---|---|
| Path Tracing NEE + MIS | Bidirectional Path Tracing | Path Tracing in Flatland | Path Density Visualization |
Scene format example:
{
"viewer" : true,
"camera" :
{
"pos" : [4, 1, -2],
"lookat": [-1, 1, 0],
"up" : [0, 1, 0],
"resolution" : [1024, 512],
"fovy": 50
},
"meshes" :
[
{
"path" : "fireplace_room/fireplace_room.obj"
}
],
"render" : [
{
"integrator" : "nextevent_path_tracer",
"output": "pt_result.pfm",
"num_spp" : 100
},
{
"integrator" : "bidirectional_path_tracer",
"output": "bdpt_result.pfm"
}
]
}
It uses its own math library and automatically switch to SIMD if your CPU supports. It strictly use double only. The performance is not its forte.