Yeah, another thing that comes to mind is image filtering. Bidir looks more blurry than normal pathtracing. Even when setting image filtering to none. !
OpenImageDenoise
Re: OpenImageDenoise
Re: OpenImageDenoise
First of all, great respect for everyone involved in this development! It was really inspiring to follow it in the forum.
I wonder that now that LuxCore has these new albedo and normal passes, maybe BCD could use them too.
I can imagine, that it would give better results in some cases than OIDN. (Probably with near final quality images.)
My idea:
I'd use the albedo pass to divide the pixels of the combined pass before denoising, and than after denoising, re-multiple the result.
I think it would help to retain the details in textures.
And I'd use the normal pass to weight the pixels with the similarity of the normals. The closer the normal of two pixels, the greater the contribution in the denoising of each other. It could help to retain the details of small geometry.
I wonder that now that LuxCore has these new albedo and normal passes, maybe BCD could use them too.
I can imagine, that it would give better results in some cases than OIDN. (Probably with near final quality images.)
My idea:
I'd use the albedo pass to divide the pixels of the combined pass before denoising, and than after denoising, re-multiple the result.
I think it would help to retain the details in textures.
And I'd use the normal pass to weight the pixels with the similarity of the normals. The closer the normal of two pixels, the greater the contribution in the denoising of each other. It could help to retain the details of small geometry.
Re: OpenImageDenoise
I've let them know on twitter...
Re: OpenImageDenoise
Ok, so first. I believe since BCD is open source project it could evolve.provisory wrote: ↑Mon Feb 04, 2019 7:25 pm First of all, great respect for everyone involved in this development! It was really inspiring to follow it in the forum.
I wonder that now that LuxCore has these new albedo and normal passes, maybe BCD could use them too.
I can imagine, that it would give better results in some cases than OIDN. (Probably with near final quality images.)
My idea:
I'd use the albedo pass to divide the pixels of the combined pass before denoising, and than after denoising, re-multiple the result.
I think it would help to retain the details in textures.
And I'd use the normal pass to weight the pixels with the similarity of the normals. The closer the normal of two pixels, the greater the contribution in the denoising of each other. It could help to retain the details of small geometry.
However... It does seem atm as a convoluted way to denoise image by gathering insane amount of histogram data and taking all the memory to process it just to throw up an ugly blurry image in double the time of actual rendering.
I'm not insisting on intel's or nvidia's solution, but BCD (in it's current state) needs to stay as far from luxcore as it is possible. It's like that idiot you don't want your daughter to date caus you know he's an idiot... And I apologize for poor comparisons, it's just the first thing that comes to mind
Re: OpenImageDenoise
I'm talking with them about BiDir: https://github.com/OpenImageDenoise/oidn/issues/10
We are all a bit surprised it doesn't work
We are all a bit surprised it doesn't work
Re: OpenImageDenoise
I really suspect image filtering. Noise/image is simply more blurry (my amateur user observation)Dade wrote: ↑Mon Feb 04, 2019 7:49 pm I'm talking with them about BiDir: https://github.com/OpenImageDenoise/oidn/issues/10
We are all a bit surprised it doesn't work
As much as I don't care about bidir, I'd like to have a full bcd replacement going on so that newcomers don't experience it
Re: OpenImageDenoise
I'm still getting unknown image pipeline error in blender on linux... Also on windows
Re: OpenImageDenoise
I tried them, the binaries are still old, I assume it is the result of MacOS build failing: https://dev.azure.com/LuxCoreRender/Lux ... uildId=198
I disabled the Azure MacOS build until when it is updated with Oidn support so it should now work: https://dev.azure.com/LuxCoreRender/Lux ... uildId=199
Re: OpenImageDenoise
Ok, will wait until they are built and retry.Dade wrote: ↑Tue Feb 05, 2019 10:51 amI tried them, the binaries are still old, I assume it is the result of MacOS build failing: https://dev.azure.com/LuxCoreRender/Lux ... uildId=198
I disabled the Azure MacOS build until when it is updated with Oidn support so it should now work: https://dev.azure.com/LuxCoreRender/Lux ... uildId=199
Re: OpenImageDenoise
Unknown image pipeline plugin type: INTEL_OIDNlacilaci wrote: ↑Tue Feb 05, 2019 10:57 amOk, will wait until they are built and retry.Dade wrote: ↑Tue Feb 05, 2019 10:51 amI tried them, the binaries are still old, I assume it is the result of MacOS build failing: https://dev.azure.com/LuxCoreRender/Lux ... uildId=198
I disabled the Azure MacOS build until when it is updated with Oidn support so it should now work: https://dev.azure.com/LuxCoreRender/Lux ... uildId=199
Just tested fresh linux build... Could this be an blender addon error maybe?