open-image-denoise を Homebrew でインストール

プロジェクトリンク

• ホームページ
• リポジトリ

コマンドとエイリアス

• oidnBenchmark
• oidnDenoise
• oidnTest

プロジェクトの歴史

The project appeared publicly as a beta in the 0.8.0 release and was promoted by Intel around GDC 2019 as an open-source denoising library for ray tracing. Intel's technical article described it as part of the Intel Rendering Framework, released under Apache 2.0, and designed to cut rendering times by filtering noise rather than requiring far more samples per pixel.

Version 1.0.0 improved quality, reduced artifacts, added memory controls for high resolutions, and made the library more practical for production integration. The 1.x series then added lightmap denoising, neural-network training code, user-trained model support, Apple Silicon work, better detail preservation, half-precision images, and the oidnBenchmark, oidnDenoise, and oidnTest example tools.

Version 2.0.0 was a major hardware expansion, adding SYCL devices for Intel Xe GPUs, CUDA devices for NVIDIA GPUs, HIP devices for AMD GPUs, asynchronous execution, device-query APIs, and graphics interop. Later 2.x releases added Metal support for Apple silicon GPUs, ARM64 CPU support, fast and high-quality modes, more GPU architectures, performance work, and fixes for device-specific issues.

採用の歴史

Open Image Denoise was designed for integration into renderers rather than as a standalone image editor. Intel's article described Unity 2019.2 lightmap integration work, and the project documentation emphasizes a flexible C/C++ API that can be incorporated into existing and new rendering solutions.

The project received a 2025 Technical Achievement Award from the Academy of Motion Picture Arts and Sciences for its contribution to the motion picture industry. That recognition is a useful adoption signal: it indicates use beyond demos, in professional rendering and production pipelines where denoising quality and predictable integration matter.

使われ方

Rendering applications feed noisy color buffers into Open Image Denoise and may also provide auxiliary albedo and normal buffers to preserve more detail. The trained RT filters are intended for both preview and final-frame rendering, covering low sample counts through nearly converged images.

Developers use the C/C++ API for embedded integration, the example tools for testing and benchmarking, and the training toolkit when renderer-specific or content-specific models are needed. The library's CPU and GPU support lets applications choose between portability, workstation acceleration, and render-farm deployment.

パッケージ好きにとっての重要性

Open Image Denoise is a good example of a package whose public CLI tools are only the tip of the dependency iceberg. The real value is a portable, permissively licensed denoising library that renderers can vendor, link, or detect at runtime.

For packagers, it is interesting because support spans CPU instruction sets, Intel/NVIDIA/AMD/Apple GPU backends, oneTBB, oneAPI/SYCL, CUDA, HIP, Metal, and external graphics-memory APIs. A simple brew formula name hides a surprisingly hardware-sensitive library.

タイムライン

• 2018: project copyright and the 0.8.0 initial beta release line appear in project materials.
• 2019: Intel promoted the library around GDC 2019 and Unity lightmap denoising work.
• 1.0.0: improved quality, memory controls, and tiled denoising support.
• 2.0.0: GPU device support expanded through SYCL, CUDA, and HIP backends.
• 2025: the Academy recognized Open Image Denoise with a Technical Achievement Award.

Related projects

• Open Image Denoise belongs to the same rendering-toolkit neighborhood as Intel Embree and OSPRay. It is commonly compared with renderer-integrated denoisers and GPU denoising APIs such as NVIDIA OptiX, but its cross-vendor CPU/GPU and Apache-licensed library model is the important packaging distinction.

ソース

• https://github.com/RenderKit/oidn
• https://raw.githubusercontent.com/RenderKit/oidn/master/README.md
• https://www.intel.com/content/www/us/en/developer/articles/technical/open-image-denoise-library-saves-time-boosts-quality.html
• https://www.openimagedenoise.org/
• https://www.openimagedenoise.org/documentation.html
• source_facts.executables
• source_facts.package-manager