This is a simple implementation of Intel’s Open Image AI denoiser. This is essentially an implementation of the example executable provided in the original repository but instead uses OIIO so that a larger variety of image formats are supports. This command line tool works the same as the Nvidia AI denoiser command line tool I made here.
Here are some examples of what it can do,
Original image:
Denoised image:
The code can be found here:
https://github.com/DeclanRussell/IntelOIDenoiser
I have created a windows distribution as well for those who wish to try it out here:
Hi Declan,
I’ve gotten your new Intel-based AI Denoiser to work. This is a 5 pass Blender render before and after applying the denoiser: https://photos.app.goo.gl/a5HiLLeqY6Q4FCvN9
Your example images above don’t really show how well it works 🙂
I had hoped I’d see similar results with high ISO photos, but this version of Intel’s Open Image Denoiser doesn’t seem to be trained for that.
Thanks again,
Dana
LikeLike
Wonderful! Glad you got it working with good results 🙂
LikeLike
Wow man it works so perfectly i did it with bad render & it make it like magic in 4 sec thanks for making this
LikeLiked by 1 person
wow thanks . i tried optiX nvidia batch script from you but it keeps failed cannot load library then tried this one and it reallly works in my i5. from now on i will use this intel denoiser
LikeLike
Hi Declan, thanks for share your implementation :). I’m wondering if there is an argument to specify the amount of denoise to use? Because I saw the examples and the denoiser looks really nice but I tried and the image still keep with some noise.
Example:
I’ve used the bump normals and refraction albedo as additional inputs with 200 repetitions but I don’t know, I suppose that it can look better maybe?
What do you think? Thanks again, cheers!
LikeLike
Hi Carlos,
Sorry for the late reply on this. Unfortunately there isn’t such an option, because the denoiser is based on machine learning its more of a “it works” or “it doesn’t work” type deal. You can improve results by improving the feature buffers that you provide the denoiser. I see you’re using the refraction albedo, why is this? Is the scene looking through some refractive object? I have the feeling that because the albedo is such a different color to the beauty it is throwing off the denoiser.
LikeLike
Hi Declan, don’t worry. I thought that can be better to denoise only one AOV instead the beauty, here is the Beauty:
As you can see the frame only have noise in the refraction AOV. So, which can be your advice? Maybe apply some green tint to te refraction albedo? Thanks!
LikeLike
Hmm I think you need a better albedo. Arnold has a dedicated albedo for denoising that deals with these kinds of things better. Which render engine are you using, does it have light path expressions or just fixed AOVs?
LikeLike
Hi Declan! Sorry for late reply. I’ve used Redshift for this render and doesn’t have LPEs or the same variety of AOVs like Arnold :/.
LikeLike
Ahh boo! I’m not very familiar with Redshift, but ideally you want some kind of albedo. I know they support the OptiX denoiser so perhaps you should ask the guys at Redshift which AOV they use for the albedo for that?
Or you could use Arnold, but I’m one of the Arnold devs, so I would say that 😛
LikeLike
Haha yes I have seen your posts in the forum, I can’t wait to go back to Arnold with GPU, right now I have more GPU power than CPU so I left Arnold the last year or more I think, but I always miss him 😂.
I’m going to investigate more about how can I use this denoiser better as you said, I can’t understand how it works so nice in the examples but here doesn’t. Thanks for your advice and your help Declan :).
LikeLike
Hello! Which one is bettter, Intel or Nvidia?
LikeLike
You tell me 🙂
In my experience the quality of the denoising varies depending on what you give either of the denoisers. Some seem to work better with some scenes than the other. As for performance though, the Nvidia denoiser wins hands down. That thing is rapid!
LikeLike
Thank you so much for this implementation! Do you think you can upgrade it to Intel’s v1.2.0? They say there’s an improvement in denoising quality
LikeLike
Sure, done!
LikeLike
You’re the best! thanks Declan!
LikeLike
Hi! Thanks for these tools.
Can yo please help what exactly I need to add to the batch script in order to utilize albedo and normal AOVs? I currently have beauty, albedo and normal AOV sequences each in separate folder. Creating a .bat file with the basic example that you provided works great, but i suppose these AOVs will give better result 🙂
LikeLike
Hey Daniel,
I didn’t add this because I couldn’t think of a nice way to deal with all the different naming conventions without the logic getting overly complicated. Perhaps this tool could help you out? http://taosoft.dk/software/freeware/dnden/
LikeLike
Hi, do you plan to update your tool with the last version of the Intel Denoiser? Thanks again for your work. Cheers. Christophe
LikeLike