scalable-coherent-path-tracing

Render engine implementation for my Master thesis subject: A Scalable and Coherent Approach to Monte Carlo Path Tracing. This thesis explores an adaptation of the Monte-Carlo path tracing algorithm that makes it more modular and scalable for GPU-based cloud environments. The proposed adaptation computes diffuse irradiance in texture space. This introduces aliasing and texture seam artifacts, which are visible in the screenshot. Solutions for these artifacts exist, but since the focus of this work was to investigate the performance of the proposed algorithm, these mitigations are not implemented here.

Setup and usage

• Make models/ directory in root folder (contains world models that will be loaded)
• Make data/world_data_textures/positions, data/world_data_textures/normals,data/world_data_textures/diffuse_coords directories in root folder (contain world data that will be loaded, and the world data will be stored here)
• Change which models will be loaded in App:loadScene()
• In app.h you can change whether the biased or unbiased approach is used
• In Renderer:initLightingTexturesPerObject and Renderer:prepareUVWorldPositionsPerObject, uncomment the lines that correspond to your model (or add new ones in case you test with new scenes)
• Make sure you pass LOAD_WORLD_DATA or CALCULATE_WORLD_DATA when calling Renderer:prepareUVWorldPositionsPerObject depending on what you desire
• In unbiasedScattering.cu and radianceCellScattering_Cubemap_hybrid.cu, make sure you uncomment the corresponding line where diffuseTexColor is set
• You need one model with lightmap UV coordinates and a copy of the same model with diffuse UV coordinates!

Algorithm overview

Overview of the proposed algorithm, which consists of three high-level passes. In the first pass direct lighting is handled via forward path tracing (A). For indirect lighting a hybrid approach is adopted in which nearby indirect radiance is path traced and distant indirect radiance is approximated. Which of the two is used is determined by a tracing range threshold parameter. Approximations are made via the use of radiance probes (small blue spheres in figure). Also notice that the scene is subdivided into so-called radiance cells. Each radiance cell computes indirect lighting for its local surfaces independently. In the second pass (B) the radiance probes are charged with indirect radiance. In the third and final pass (C), the surfaces local to a radiance cell are illuminated by incoming indirect radiance. Passes B and C can be repeated based on the amount of lighting bounces that the user desires.

Screenshot