Intel Open Image Denoise
Intel Open Image Denoise Takes Away the Award for Scientific and Technical Achievement
A Technical Achievement Award will be given by the Academy of Motion Picture Arts and Sciences to Intel Open Image Denoise, an open-source library that offers a high-performance, high-quality AI-based denoising solution for ray traced images. The Academy, which organizes the yearly Oscars, acknowledged the library as a pioneer in contemporary filmmaking.
Ray tracing is the cornerstone of contemporary rendering. Although the algorithm is strong and can create incredibly lifelike visuals, it requires a lot of processing power. Ray tracing alone requires tracing a lot of rays, which is frequently time-consuming and costly, in order to create images free of noise and artefacts. It is possible to trace fewer rays without compromising image quality by augmenting the renderer with a competent denoiser, such as Intel Open Image Denoise, which can drastically cut down on rendering times.
The purpose of Intel Open Image Denoise is to reduce rendering times and speed up real-time previews throughout the creative process by employing AI neural networks to filter out the undesired noise that comes with ray tracing. Its straightforward yet adaptable C/C++ API makes it easy to incorporate the library into the majority of new or current rendering solutions. It also supports a wide range of cross-vendors, including optimisations for the majority of the main CPU and GPU architectures from Apple, AMD, Nvidia, Arm, and Intel.
Intel Open Image Denoise operates under the Apache 2.0 license and is a component of the Intel Rendering Toolkit. The widely used, highly effective, and detail-preserving U-Net architecture provides its basic technology, improving the industry’s standard for computer-generated imagery. More flexibility and better image quality are possible because the library is free and open source, and the training toolkit that comes with it allows users to train custom denoising models using their own datasets. Additionally, the integrated denoising neural networks can be re-trained for users’ own renderers, styles, and films by producers and companies.
Effective deep learning-based denoising filters that can handle a broad range of samples per pixel (spp), from 1 spp to nearly entirely converged, are the foundation of the Intel Open Image Denoise library. It is therefore appropriate for rendering both previews and final frames. Images can be denoised by the filters using just the noisy colour (beauty) buffer or, if desired, auxiliary feature buffers (e.g. albedo, normal) to retain as much detail as possible. The majority of renderers provide such buffers as arbitrary output variables (AOVs), or they are typically easy to accomplish.
Although a collection of pre-trained filter models is included with the library, using them is not required. It is able to train the model with the included training toolkit and user-provided picture datasets to optimise a filter for a particular renderer, sample count, content type, scene, etc.
The CPUs and GPUs that Intel Open Image Denoise supports range widely from various vendors:
- CPUs that support Intel 64 architecture (with at least SSE4.1)
- Apple silicon CPUs are ARM64 (AArch64).
- Dedicated and integrated GPUs for the Intel Xe, Xe2, and Xe3 architectures include Intel Arc B-Series Graphics, Intel Arc A-Series Graphics, Intel Arc Pro Series Graphics, Intel Data Centre GPU Flex Series, Intel Data Centre GPU Max Series, Intel Iris Xe Graphics, Intel Core Ultra Processors with Intel Arc Graphics, 11th–14th Gen Intel Core processor graphics, and associated Intel Pentium and Celeron processors.
- NVIDIA GPU architectures include Volta, Turing, Ampere, Ada Lovelace, Hopper, and Blackwell.
- AMD GPUs with RDNA2 (Navi 21 only), RDNA3 (Navi 3x), and RDNA4 chips
- Apple silicon GPUs, including the M1 model
Most computers, including laptops, workstations, and compute nodes in high-performance computing systems, can execute it. Because of its efficiency, it can be used for interactive or even real-time ray tracing, depending on the technology, in addition to offline rendering.
Intel Open Image Denoise achieves great denoising performance by taking advantage of the tensor cores on NVIDIA GPUs, the Intel Xe Matrix Extensions (Intel XMX) on Intel GPUs, and the current instruction sets SSE4, AVX2, AVX-512, and NEON on CPUs.
Intel Open Image Denoise System Specifications
An Intel 64 (with SSE4.1) or ARM64 architecture compliant CPU is required to run Intel Open Image Denoise, together with a 64-bit version of Windows, Linux, or macOS.
Additionally, please install the most recent Intel graphics drivers for support of the Intel GPU:
- Windows: Intel Graphics Driver 31.0.101.4953 or newer
- Linux: Intel software for General Purpose GPU capabilities release 20230323 or newer
Intel Open Image Denoise may operate with limited capabilities, perform less than optimally, or become unstable if you use outdated driver versions. Resizable BAR is also required in the BIOS for Intel dedicated GPUs while operating on Linux, and it is highly advised when operating on Windows.
Also, make sure you have the most recent NVIDIA graphics drivers installed for NVIDIA GPU support:
- Windows: Version 528.33 or newer
- Linux: Version 525.60.13 or newer
Please additionally install the most recent AMD graphics drivers in order to support AMD GPUs:
- AMD software on Windows (Adrenalin Edition 25.3.1 or later)
- Linux: Radeon Software for Linux, up to version 24.30.4
A macOS version of Ventura or later is necessary for Apple GPU compatibility.