exclude Alpha from being denoised

[Wird]

I would really appreciate if there is an option to exclude Alpha from being denoised, as it causes some really ugly artifacts under certain circumstances.
The ideal solution would be to have the option for both denoised and orginal Alpha at once.

[hermit.crab]

Refer to this thread for a solution:
https://forums.chaosgroup.com/forum/v-ray-for-3ds-max-forums/v-ray-for-3ds-max-problems/1041763-both-denoiser-change-alpha-channel?p=1042013#post1042013

[Wird]

Hello Aleksandar,
yes, but this solution does not work form me. In my workflow I use PNGs and Photoshop (usually 8 or 16 bit).
As I understand - I would be forced to use EXR as the render output, because otherwise the "denoise" elements are not saved as separate output channels in PNG, right? (but why?)
And Photoshop does not read multi-channel EXRs, or does it?

[Codi]

And Photoshop does not read multi-channel EXRs, or does it?

It can read EXRs using ProEXR free PS plugin: http://www.fnordware.com/ProEXR/

[vaya_studio]

Or you can use this one as well: https://www.exr-io.com/
It's also free and has some pretty neat features.

[^Lele^]

As I understand - I would be forced to use EXR as the render output, because otherwise the "denoise" elements are not saved as separate output channels in PNG, right? (but why?)

Because EXR has the right bit depths (16 and 32 bpc float) needed for the denoiser to perform its maths, which PNG lacks (so the written data would essentially be garbage).
EXR's been a staple of CG for over 20 years now, i'd make it a priority to figure out what they are about, and conform my workflow to them.
Literature is ample, easy to come by, and not all that difficult to understand from a practical standpoint (as it solves a number of outstanding issues everyone faces with lower bit depths.).

[Vizioen]

Too bad one of the most used image editing programs doesn't support it natively and then some. I guess that's why a lot of artists (that aren't into compositing software) are still reluctant to use EXR, even here in the studio I've been trying to push EXR as a standard but they still love TGA so much, which I detest.

[Wird]

Hello,
thanks for your outlook. I'm aware of 32bit editing capabilities/advantages, but I work mainly on still images in Photoshop - so the 8/16 bit images are my choice (unfortunatelly perhaps...)

Regarding this - Is there any option/utility (perhaps even 3rd party solution) how to automatically extract 8/16 bit PNGs out from the multi-channel EXRs?
That would help me a lot.

[^Lele^]

Hello,
thanks for your outlook. I'm aware of 32bit editing capabilities/advantages, but I work mainly on still images in Photoshop - so the 8/16 bit images are my choice (unfortunatelly perhaps...)

Regarding this - Is there any option/utility (perhaps even 3rd party solution) how to automatically extract 8/16 bit PNGs out from the multi-channel EXRs?
That would help me a lot.

You are confusing bitness types.
PNGs, and TIFF can do 8 and 16 bit INTEGER data (say, a channel can have a pixel of value 254 or 255, or 64534 or 64535, but not fractions of that).
EXRs (and other such formats, like vrimage and tx) can do 16 and 32 bpc FLOAT, where white is 1.000, and values can smoothly (instead of stepping) vary above and below that value (0.999 and 1.001 are valid, for example.).
That in turn leads to more accurate maths for all the DATA channel types (normals, z-depth, pPos, velocity, and so on), never mind the beauty layers.

Conversion to 16 bit integer of 16 (or 32) bit float data will result in loss of precision, and render those element useless to the accurate math's needs (i'm aware one can botch their way through a normal pass in PS. hence me mentioning accurate.), like that of the denoiser.

Said conversion is fine once one has finished the accurate processing part, and needs either publishing, or more artistic work done (f.e. in PS if they so choose. i wouldn't, but it's admittedly less critical than with data elements.)

To achieve that, there are a number of tools which are free or very cheap (fusion, affinity, natron, to name a few with solid Floating Point maths, along with a few tools which allow for thumbnail display in windows explorer and so on.), so there really isn't a valid excuse to not use the format, and all it brings to the table, as i see it, considering that pre- and post-processing steps after the render phase have also been the norm for two decades, and not just in big VFX studios...