Shilin Zhu, Mark Meyer
Abstract:
Denoising path traced images is essential in the production rendering pipeline. Existing denoisers only apply to 2D flat images, which introduces challenges in the compositing stage where multiple rendered components are combined together to produce the final look. In recent years, deep image has become more popular and preferable in the industry because it can store values at different depths, making deep compositing possible. Despite the benefit of using a deep image, the lack of a proper denoising algorithm introduces issues in production, and the standard flat image denoiser is not directly applicable by design because of its inability to process depth bins. In this project, we develop a compatible 3D denoiser that can process deep pixels effectively, which opens up more possibilities in post processing such as compositing and image editing.
Paper (PDF)
Technical Report