NVIDIA AI denoiser upscaler

[bos]

Hello, I'm just testing out the latest update and I am getting a odd result when I enable the AI upscaler. Am I setting this up wrong? It splits the render into this while it renders but when its done it stays like this and doesn't merge the results.

[matanov]

From the docs:

NVidia AI upscale – When enabled, the rendered image is internally calculated with half the resolution set in the render setup (in each dimension), and the denoised image is upscaled to the full resolution using NVIDIA's AI upscaling technology. NVIDIA's AI upscaling technology represents an advanced method based on machine learning which provides better and more detailed results when compared to a standard image interpolation. For example, a 3840x2160 resolution renders the image internally in 1920x1080 and the AI upscaling denoiser produces the final 3840x2160 image from the lower resolution render.

https://docs.chaos.com/display/VMAX/VRayDenoiser

[matanov]

Hello,

"Temporal mode" doesn't work with upscaling (this is not on our side). It will be turned off automatically in the future builds

p.s. mind that such features also rely on the GPU driver, so that can also affect the result

[bos]

Ok, I did just update to the most recent studio driver. Nvidia 528.49. Maybe that's part of it. Cool feature excited to test it more.

[jackieteh]

NVIDIA AI denoiser upscaler?

I am still not understand what is this feature, could anyone can explain further?

Thanks in advance

[matanov]

From the docs:

NVidia AI upscale – When enabled, the rendered image is internally calculated with half the resolution set in the render setup (in each dimension), and the denoised image is upscaled to the full resolution using NVIDIA's AI upscaling technology. NVIDIA's AI upscaling technology represents an advanced method based on machine learning which provides better and more detailed results when compared to a standard image interpolation. For example, a 3840x2160 resolution renders the image internally in 1920x1080 and the AI upscaling denoiser produces the final 3840x2160 image from the lower resolution render.

https://docs.chaos.com/display/VMAX/VRayDenoiser

[jackieteh]

From the docs:

NVidia AI upscale – When enabled, the rendered image is internally calculated with half the resolution set in the render setup (in each dimension), and the denoised image is upscaled to the full resolution using NVIDIA's AI upscaling technology. NVIDIA's AI upscaling technology represents an advanced method based on machine learning which provides better and more detailed results when compared to a standard image interpolation. For example, a 3840x2160 resolution renders the image internally in 1920x1080 and the AI upscaling denoiser produces the final 3840x2160 image from the lower resolution render.

https://docs.chaos.com/display/VMAX/VRayDenoiser

Hi, your link is broken, i cannot visit the page

anyway, i did read the vray document page "For example, a 3840x2160 resolution renders the image internally in 1920x1080 and the AI upscaling denoiser produces the final 3840x2160 image from the lower resolution render." where can I set the size? or i just set the 3840x2160, vray will automatically calculate in 1920x1080 and also upscale it to 3840x2160 automatically?

Will this feature help speed up rendering time?

[matanov]

Hi, your link is broken, i cannot visit the page

For some reason the forum broke the link. It should be fixed now.

or i just set the 3840x2160, vray will automatically calculate in 1920x1080 and also upscale it to 3840x2160 automatically?

Will this feature help speed up rendering time?

Exactly. Rendertime can be down to 4X because V-Ray is rendering 4 times less pixels. Don't expect miracles, though. It generally works quite ok for quick previews, but it might lead to loss of quality in some cases.

[seandunderdale]

I saw some tests at 1000 x 1000 online, and clearly the effect works, but suffers at low resolutions. Would it be correct to think the bigger you render, the more useful this tool is? If you render a 10k image for example, a 5k downsize render would still contain a lot of fidelity. Im super excited to try this out on some high resolution automotive renders.

[matanov]

I saw some tests at 1000 x 1000 online, and clearly the effect works, but suffers at low resolutions. Would it be correct to think the bigger you render, the more useful this tool is? If you render a 10k image for example, a 5k downsize render would still contain a lot of fidelity. Im super excited to try this out on some high resolution automotive renders.

Yes, but there always these parts of the image where you need the actual pixels

[jackieteh]

For some reason the forum broke the link. It should be fixed now.




Exactly. Rendertime can be down to 4X because V-Ray is rendering 4 times less pixels. Don't expect miracles, though. It generally works quite ok for quick previews, but it might lead to loss of quality in some cases.

Good for preview