CUDA Denoiser For CUDA Path Tracer

University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4

• Shenyue Chen
  • LinkedIn, personal website
• Tested on: Windows 10, Intel Xeon Platinum 8259CL @ 2.50GHz 16GB, Tesla T4 (AWS g4dn-xlarge)

Performance analysis

Render time

For the cornell box scene, we measure the average time per iteration, which is quite stable, around 0.11s per iteration.

Denoise time

To analyze the time used to denoise, first we consider the image of 800 * 800 resolution, we change the number of filter levels. We could observe that the time needed to denoise increases for the first levels but remain stable when the stepwidth gets too big, which is expected. Compare the denoise time with the render timer per iteration we get from the previous section. The denoise time is roughly equal to one iteration for the cornell box scene.

Denoise performance for different levels

We can observe the denoise performance for different levels under the same cornell scene. The result clearly gets better as the number of filter levels grow.

200 iter 1 filter levels

200 iter 3 filter levels

200 iter 5 filter levels

Denoise in different resolutions

To measure the denoise time in different resolutions, we very the resolution from 400 * 400 to 1000 * 1000, all using 5 denoise levels. From the plot we can see that the denoise time increases in a linear manner.

Denoise performace in different scenes

We can compare the performance of the denoiser in different scenes. Here we choose 50 iterations + 5 filter levels. From the results, it seems that the denoiser performs better in fully-lighted scenarios.

	Original	Denoised
Cornell
Cornell celing