Archive notice: This article was originally published on February 1, 2019. Links and embedded videos are preserved as part of the historical record.
Intel already announced at SIGGRAPH 2018 that it was developing a new Open Image Denoise library. In the course of this, the company also joined the Academy Software Foundation . Now the source code for the project has been published on Github: Open Image Denoise thus expands the Intel Rendering Framework, which also includes the Embree kernels, the raytracing renderer OSPRay and the software rasterizer OpenSWR.
Intel Open Image Denoise
As with other well-known denoisers, the software filters noise from rendered images created with raytracing methods, thus reducing the number of required samples per pixel. Open Image Denoise is internally based, among other things, on the Math Kernel Library for Deep Neural Networks (MKL-DNN). Like the well-known GPU counterpart OptiX from Nvidia, the system therefore uses deep-learning algorithms to improve denoising. Intel has placed example test runs in this gallery – however, there is still no more precise data on the speed of the denoising process.
Requirements & Download
Unlike OptiX, however, Intel’s denoiser is also intended to be compatible with CPUs that are not made by Intel itself. The requirements for Windows, Linux or Mac are a CPU with SSE4.2 support and a 64-bit architecture – AMD processors would therefore also be compatible. The already compiled Open Image Denoiser is currently available under an Apache 2.0 licence in the Beta version 0.8 here and can also be used independently of Intel’s Rendering Framework. The source code on Github can be found here.
Further Information: To the OIDN Website
