Intel releases open-source denoiser Open Image Denoise
Intel has made the source code for Open Image Denoise, its new open-source CPU-based render denoising system, available on GitHub.
As with Nvidia’s popular GPU-based OptiX denoising technology, the system uses machine learning techniques to accelerate the process of removing noise from rendered images.
An AI-driven, CPU-based, hardware-agnostic render denoiser
First announced at Siggraph 2018, Open Image Denoise is a set of “high-performance, high-quality denoising filters for images rendered with ray tracing”.
It’s designed to remove the noise generated by the Monte Carlo rendering techniques like path tracing, widely used in modern production renderers.
However, it does build on MKL-DNN, Intel’s Math Kernel Library for Deep Neural Networks, meaning that like OptiX – Nvidia’s popular GPU-based denoising technology, integrated into many production renderers – it uses AI techniques to accelerate denoising.
Unlike OptiX, Open Image Denoise isn’t hardware-specific: while it’s designed for Intel 64 CPUs, it supports “compatible architectures”, although we aren’t sure how well it runs on AMD processors.
So how well does it work on real production scenes?
Intel’s original demo showed Open Image Denoise in use with the multi-billion-object Moana island data set, which Disney made freely available for research purposes last year.
For those with slightly more modest production requirements, the gallery section of Intel’s product website shows it in use with standard Lumberyard, Crytek and Evermotion test scenes.
This thread on BlenderArtists.org also shows it in use on automotive and interior shots, suggesting that output quality compares favourably to Blender’s standard denoiser.
Availability and system requirements
Open Image Denoise is available for 64-bit Windows, Linux and macOS under an Apache 2.0 licence. A CPU with support for at least SSE4.2 is required to run it.
Read more about Open Image Denoise on Intel’s product page
(Includes links to downloads)