AI DeNoiser

[Allan]

I am not a fan of Solidworks Visualizer, but this AI DeNoiser looks amazing. It promises 10x increased rendering speeds.
Can you make something similar in Keyshot?



Source:
http://blogs.solidworks.com/solidworksblog/2018/02/solidworks-2019-technology-preview-part-1.html

[KeyShot]

We have been looking into this. One of the issues is that it is not a general solution. It will not work well with certain types of noise. It also tends to overblur certain regions (as can be seen in your example). One of the reasons why GPU renders use (or have to use) this type of technique is that they generally all use some sort of path tracing. Path tracing is a fast simple Monte Carlo sampling technique that GPUs are able to run. Path tracing will give you a noisy image fast, but it will take a very long time to get rid of the noise. Mathematically, if you wish to half the amount of noise you will need four times as many samples (and four times the rendering time). This makes it very difficult to achieve a noise free image, and this is the reason why path tracing is being paired with a noise filter (I even wrote a simple paper about this long time ago: http://graphics.stanford.edu/~henrik/papers/wscg95.html).
In KeyShot we have focused on reducing noise as part of the rendering. It is a bit more complex, but something we can do, since we do not use path tracing, but a more complex algorithm. When you using interior mode there is already an adaptive denoising taking place. We are still refining this. There are still cases where hot pixels can appear, and we have some work in progress to specifically address those.

[Allan]

Thanks for the reply. Yes, it is probably not so straightforward to work with. I am glad that you work on the hot pixels. That is one of my main issues with Keyshot at the moment.

[mattjgerard]

I've found that the Nik denoiser works wonders for my images. And it will process it as a seperate layer, so things can be masked out and different levels of denoise can be applied to different areas of the image. Very flexible, but yes it takes time in post to do it.

The adaptive part of the noise reduction being built into keyshot itself seems to be the best news for this.

[PerFotoVDB]

Hi

A built-in denoiser would be awesome!

Would speedup render times tremendously.

Cheers

[mafrieger]

thanks for starting discussing this. Was the same thing this post was about:
Look into Denoising at GPU: to receive ok quality much faster
https://www.keyshot.com/forum/index.php?topic=21347.0

[RRIS]

I want to ask, is it worth it to invest in a denoiser for animation? So, batch processing a lot of images without manual edits?

I'm looking into Denoise from Topaz, thinking it might be more efficient to increase render resolution a little with more noise.. then denoise and resize to final resolution.
Has anyone tested this?

(Just found this btw, a command line app based on nVidia's AI denoiser: https://declanrussell.com/portfolio/nvidia-ai-denoiser/ )

[KeyShot]

We have something coming in KeyShot, so if you can be a little patient then you do not have to invest in it. There is a denoiser today in interior mode, but we will add more general support for denoising.

[mafrieger]

Maybe this is interesting for Keyshot makers:

intel released an CPU based denoiser with Open Souce Licence (Apache 2.0)
It's supporting all CPUs with SSE4.2 and above.
latest version: 0.8 beta

Intel® Open Image Denoise

This high-performance, open-source library improves visual quality during interaction by using machine learning methods (including functions in Intel® Math Kernel Library for Deep Neural Networks) to selectively filter noise. This independent component can be used for noise reduction on 3D rendered images, with or without Intel Embree.

More Informations should be available soon at https://software.intel.com/en-us/rendering-framework

[KeyShot]

Thank you. Yes, we have seen this and are investigating it.

[Allan]

Thank you. Yes, we have seen this and are investigating it.

Exciting!