Intel Open Image Denoise Wins Scientific and Technical Achievement Award

[btarunr]

Tomorrow, the Academy of Motion Picture Arts and Sciences will honor Intel Open Image Denoise, an open source library that provides a high-performance, high-quality AI-based denoising solution for ray traced images, with a Technical Achievement Award. The Academy, which hosts the annual Oscars, recognized the library as a contributing innovation in modern film rendering.

"Intel continues to push the envelope on high-quality rendering and visual experiences, delivering benefits to creators around the world, and this achievement is a timely recognition for the passion and persistence of the team. We are committed to bringing the best visual computing breakthroughs for creators, developers and gamers with our products and software ecosystem, and this top award is a manifest of that coming to life," said Anton Kaplanyan, Intel vice president, Graphics and GPU Research.



Ray tracing is the foundation of modern rendering. It is a powerful algorithm capable of producing very realistic images, but it has high computational requirements. When using ray tracing on its own to produce noise- and artifact-free images, a large number of rays need to be traced, which is often slow and expensive. With a sophisticated denoiser like Intel Open Image Denoise, rendering times can be significantly reduced by augmenting the renderer, which enables tracing fewer rays without sacrificing image quality.

Intel Open Image Denoise filters out the unwanted noise inherent to ray tracing using AI neural networks, accelerating real-time previews during the creative process and cutting down final rendering times. Due to its simple but flexible C/C++ API, the library can be easily integrated into most existing or new rendering solutions. It also has rich cross-vendor support, with optimizations for most major CPU and GPU architectures from Intel, AMD, NVIDIA, Apple and Arm.

Intel Open Image Denoise is part of the Intel Rendering Toolkit and is released under the Apache 2.0 license. Its core technology is provided by the widely adopted U-Net architecture that is highly efficient and preserves details, raising the quality of computer-generated imagery across the industry. Since the library is open source and free to use, the included training tool kit enables users to train custom denoising models using their own datasets, allowing for more flexibility and higher image quality. Additionally, creators and studios can also re-train the included denoising neural networks for their own renderers, styles and films.

The library is also integrated into many well-known rendering tools, including Autodesk Arnold, Blender, Chaos V-Ray, Chaos Corona, Foundry Modo, Maxon Cinema 4D and others.

The Scientific and Technical Awards were introduced in 1931 to recognize the achievements in science and technology's critical role in advancing motion picture production. The Academy will honor scientific and technical achievements at the annual Scientific and Technical Awards ceremony on April 29.

View at TechPowerUp Main Site

 

[Chaitanya]

This was kind of use cases that we should have seen more off rather than the spywares we have been delivered on a silver platter.

 

[unwind-protect]

Well deserved.

Intel put out some good software and libraries. Which I use on my AMD CPUs, eh

 

[ZoneDymo]

ok so what does this mean? several games have RT in it with some build in denoiser, does the game need to be patched to support this denoiser if found superior? is it something a gpu just does by itself?

 

[Random_User]

IMHO, all the similar tech (and in general), should be open to the publick. The more it's accessible, the more use it has, and thus more developed and HW-neutral it becomes.
But, but... then, it would force the GPU makers to put more efforts into R&D, as it will show who is the "s*cker", and is a lazy scammer.

 

[Visible Noise]

ok so what does this mean? several games have RT in it with some build in denoiser, does the game need to be patched to support this denoiser if found superior? is it something a gpu just does by itself?

This isn’t for games. RTFA.

 

[R-T-B]

This isn’t for games. RTFA.

This is really ironic considering the article literally mentions PC ray tracing as a use case.

 

[MentalAcetylide]

I guess this should be much better than the denoiser I've used in the past in conjunction with Iray rendering. I've always had the problem of it either removing too much detail, reduced image crispness, or both.

 

[Scrizz]

This is really ironic considering the article literally mentions PC ray tracing as a use case.

PC ray tracing has been around decades before it came to games.
Ray tracing can be in games, but ray tracing isn't games.

 

[R-T-B]

PC ray tracing has been around decades before it came to games.
Ray tracing can be in games, but ray tracing isn't games.

Regardless, I'm pretty sure denoising being applied to ray tracing for the purpose of real-time rendering has a pretty obvious main use case here, and its games or at mininum, things that are realtime. You train these neural nets to get real time results.

If time is not a constraint you don't need a denoiser.

 

[dyonoctis]

Regardless, I'm pretty sure denoising being applied to ray tracing for the purpose of real-time rendering has a pretty obvious main use case here, and its games or at mininum, things that are realtime. You train these neural nets to get real time results.

If time is not a constraint you don't need a denoiser.

Real-time most likely refers to viewport rendering, or iterative rendering. Which are "real-time" because they refresh the render at each modification. It's often used in the lookdev stage. Those offline 3D denoiser aren't nimble enough for games, even Nvidia isn't using their own Optix for games, they are using a different set of technologies. Both optix and Open image are available in UE 5, but it's really more for prerendered sequence/look dev than gameplay. Ray reconstruction is a thing because denoisiers for gameplay have limitation linked to the speed at which they are required to operate, and the very low sample that RT games have to be rendered with.
NVIDIA OptiX™ AI-Accelerated Denoiser | NVIDIA Developer

A denoiser still has its place for a final render: depending on what you do, the number of samples you would need to add to get a clean image wouldn't be worth the trade, especially for animation. Even the like of pixars are spending R&D into developing a production-ready denoiser.

Denoiser - RenderMan - RenderMan 26

 

[Visible Noise]

Regardless, I'm pretty sure denoising being applied to ray tracing for the purpose of real-time rendering has a pretty obvious main use case here, and its games or at mininum, things that are realtime. You train these neural nets to get real time results.

If time is not a constraint you don't need a denoiser.

There’s nothing that says this is real-time. Look at the listed applications, none of those are real-time. Also, yes, off-line rendering still uses denoisers.