Denoising and guided upsampling of Monte Carlo pathtraced low resolution renderings

Download
2021-2-15
Alpay, Kadir Cenk
Monte Carlo path tracing is used to generate renderings by estimating the rendering equation using the Monte Carlo method. An extensive amount of ray samples per pixel is needed to be cast during this rendering process to create an image with a low enough variance to be considered visually noise-free. Casting that amount of samples requires an expensive time budget. Many studies focus on rendering a noisy image at the original resolution with a decreased sample count and then applying a post-process denoising to produce a visually appealing output. This approach speeds up the rendering process and creates a denoised image of comparable quality to the visually noise-free ground truth. However, the denoising process cannot handle the noisy image’s high variance accurately if the sample count is decreased harshly to complete the rendering process in a shorter time budget. In this thesis work, we try to overcome this problem by proposing a pipeline that renders the image at a reduced resolution to cast more samples than the harshly decreased sample count in the same time budget. This noisy low-resolution image is then denoised more accurately, thanks to having a lower variance. It is then upsampled with the guidance of the auxiliary scene data rendered swiftly in a separate rendering pass at the original resolution. Experimental evaluation shows that the proposed pipeline generates denoised and guided upsampled images in promisingly good quality compared to denoising the noisy original resolution images rendered with the harshly decreased sample count.
Citation Formats
K. C. Alpay, “Denoising and guided upsampling of Monte Carlo pathtraced low resolution renderings,” M.S. - Master of Science, Middle East Technical University, 2021.