Skip to content

zhengliii/Project4-CUDA-Denoiser

 
 

Repository files navigation

CUDA Denoiser For CUDA Path Tracer

University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 3

  • Li Zheng
  • Tested on: Windows CUDA10, i5-3600 @ 3.59GHz 16GB, RTX 2060 6GB (personal computer)

Performance Analysis

  • How denoising influences the number of iterations needed to get an "acceptably smooth" result?
    For the cornell_ceilight_light scene, the denoising takes 10 iterations to smooth the rendering result. Without denoising, it takes about 600 iterations to get an "acceptably smooth" result.

  • How it impacts runtime at different resolutions?

    The runtime increases with resolution. Approximately, runtime is in proportion to resolution.x * resolution.y.

  • How effective/ineffective it is with different material types?
    It works well with diffusive material, but is ineffective with reflective and refractive material.

Visual results and performance for varying filter sizes (10 iterations)

Filter size 10 30
Visual result
Time(ms) 37.6 50.1
Filter size 60 80
Visual result
Time(ms) 67.0 105.5

Visual results of different scenes

small light large light

The scene with large light has a better performance. Because the traced rays have higher probability to hit the large lights and terminates earlier, which introduces less noise.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 85.4%
  • C 13.1%
  • Other 1.5%