LRC AI Denoise DNG files

If you would like to post, you'll need to register. Note that if you have a BCG store account, you'll need a new, separate account here (we keep the two sites separate for security purposes).

bgwilliams47

bgwilliams47

New member
Member's gallery
LRC AI Denoise is producing DNG files with bit depth of 24 (according to Windows File Explorer - properties) i.e. an 8 bit resolution per band.
My Sony Alpha has a 14-bit ADC, so I guess my raw files have 14 bits per band.
Both lightroom and photoshop can process with 16 bits per band

According to this link, AI denoise works on the RAW files.

This looks like throwing away 6 stops of detail, greatly comprising the ability to extract details from shadows or highlights.

Does anyone have relevant insights or experience?
 
One thing I do know is that the camera photosites/pixels are monochrome, each filtered to either red or green or blue. So the raw file is not per channel, just 14 bit per pixel. The RGB color is created by the raw converter, looking at a pixel of a single color but also using info from neighboring pixels to fill in the other two colors. Only after it gets demosaiced by the raw converter does it then have 3 channels each at whatever bit depth the raw converter uses, I believe lightroom internally is working at 16 bit per channel, but this is not baked in until you export. You can export a tiff at 16 bit per channel if set that way in preferences.

So I'm wondering, since the dng is still considered a raw file, if you are really not throwing away any bits after all. I sure don't see any diminished output between the two.
 
Last edited:
Member's gallery
bgwilliams47 said:
LRC AI Denoise is producing DNG files with bit depth of 24 (according to Windows File Explorer - properties) i.e. an 8 bit resolution per band.
My Sony Alpha has a 14-bit ADC, so I guess my raw files have 14 bits per band.
Both lightroom and photoshop can process with 16 bits per band

According to this link, AI denoise works on the RAW files.

This looks like throwing away 6 stops of detail, greatly comprising the ability to extract details from shadows or highlights.

Does anyone have relevant insights or experience?
Click to expand...

You're confusing bit-depth (color) with dynamic range. You're not losing any details in shadows or highlights.
 
Member's gallery
Nimi said:
You're confusing bit-depth (color) with dynamic range. You're not losing any details in shadows or highlights.
Click to expand...
Thanks Nimi,
I have seen and fully understand that video.

Bit-depth is directly related to dynamic range, not confused with it.
8 bit data has 256 different values
14 bit data has 16,384 values

So, for example with 14 bit data much have some low light detail in the range 0-63, which can be recovered by tone adjustments.
When the 14 bit data is converted to 8 bit, that range is collapsed to a single value, 0, and all detail is lost.

Under perfect circumstances (good lighting, correct exposure, uncomplicated subject etc) 8 bit data will be adequate. Then loss of dynamic range won't be obvious.

However, for imperfect situations, more bits are used:-
  • Phones have 10 or 12 bit ADCs (analogue to digital converters);
  • Cameras that I know of have 12 or 14 bit ADCs
  • Lightroom and Photoshop use (optionally) 16 bit data with Prophoto RGB.
I remain concerned that LRC AI denoise may not be 100% reliable.
 
Here is a good example of how bit depth affects dynamic range. On the Canon R5 the mechanical shutter is 14 bit, but the electronic shutter is 12 bit. As you see from the chart, they are equal from iso 800 and higher and similar starting around iso 400. Only below iso 400 is any real difference in dynamic range. The reason has to do with how the floor for dynamic range is defined as where noise and signal can't be distinguished from each other. In other words the noise floor is where dynamic range starts. So yes it does matter, but only if shooting close to base ISO.

Beyond that you can't really conflate the bit depth of the photoshop or lightroom RGB with the capability of the sensor to parse shadow detail.

 
Last edited:
Member's gallery
bleirer said:
One thing I do know is that the camera photosites/pixels are monochrome, each filtered to either red or green or blue. So the raw file is not per channel, just 14 bit per pixel. The RGB color is created by the raw converter, looking at a pixel of a single color but also using info from neighboring pixels to fill in the other two colors. Only after it gets demosaiced by the raw converter does it then have 3 channels each at whatever bit depth the raw converter uses, I believe lightroom internally is working at 16 bit per channel, but this is not baked in until you export. You can export a tiff at 16 bit per channel if set that way in preferences.

So I'm wondering, since the dng is still considered a raw file, if you are really not throwing away any bits after all. I sure don't see any diminished output between the two.
Click to expand...
Thanks Bill,
An interesting point.
This link https://rawpedia.rawtherapee.com/How_to_convert_raw_formats_to_DNG says:
"The DNG format can contain real raw data, but it can also contain demosaiced images. These demosaiced images are no longer really raw - they have been pre-cooked. While this is generally undesirable, there are situations where we can take advantage of this possibility. "​
I think that de-noising will include demosaicing. Anyway, in this case the DNG is not a raw file, it has been demosaiced.

BUT I WAS WRONG. File explorer reported bit depth of 24 & I concluded 8 bits per pixel.
Now when LRC examines the same file, it reports 14 bits.
SO file explorer can't handle DNG metadata.
CASE CLOSED

1715219523683.png
 
Member's gallery
bleirer said:
Here is a good example of how bit depth affects dynamic range. On the Canon R5 the mechanical shutter is 14 bit, but the electronic shutter is 12 bit. As you see from the chart, they are equal from iso 800 and higher and similar starting around iso 400. Only below iso 400 is any real difference in dynamic range. The reason has to do with how the floor for dynamic range is defined as where noise and signal can't be distinguished from each other. In other words the noise floor is where dynamic range starts. So yes it does matter, but only if shooting close to base ISO.

Beyond that you can't really conflate the bit depth of the photoshop or lightroom RGB with the capability of the sensor to parse shadow detail.

Click to expand...
I should have said that bit-depth in a file represent potential or maximum dynamic range. Actual dynamic range will depend on the statistics of your image.
I now believe that I was mislead by Windows File Explorer - properties misreporting bit-depth of DNG files.
 
bgwilliams47 said:
Thanks Bill,
An interesting point.
This link https://rawpedia.rawtherapee.com/How_to_convert_raw_formats_to_DNG says:
"The DNG format can contain real raw data, but it can also contain demosaiced images. These demosaiced images are no longer really raw - they have been pre-cooked. While this is generally undesirable, there are situations where we can take advantage of this possibility. "​
I think that de-noising will include demosaicing. Anyway, in this case the DNG is not a raw file, it has been demosaiced.

BUT I WAS WRONG. File explorer reported bit depth of 24 & I concluded 8 bits per pixel.
Now when LRC examines the same file, it reports 14 bits.
SO file explorer can't handle DNG metadata.
CASE CLOSED

View attachment 88617
Click to expand...

It makes sense as you say that the dng would have to be demosaiced in order to do the enhanced noise reduction. So are you still concerned about lost bit depth or satisfied that it is all good?
 
JoelKlein said:
Is NX Studio also demosaiced after running noise reduction?
Click to expand...

No, it would have to create a new file for that image to do that. It only bakes in the changes it makes when saving a tiff, jpeg, etc.

All of these raw converter programs temporarily render the raw mage into an rgb image so we can work on them, but those changes are not baked in and stay fluid until a new file type is created with export or save as.
 
You got me curious. This is quoted from the linked article:

Denoise uses machine learning to interpolate those patterns and remove noise at the same time. That is, our models are designed and trained to perform both demosaicing and denoising in a single step.

blog.adobe.com

Denoise Demystified | Adobe Blog

Get a closer look at Denoise, our latest AI-powered feature for Lightroom and Camera Raw, explained by Adobe Fellow and ACR team member, Eric Chan. “From the ACR team” is a blog series that brings you insights directly from the team that builds the imaging features for Lightroom, Lightroom...
blog.adobe.com
 
Member's gallery
bleirer said:
No, it would have to create a new file for that image to do that. It only bakes in the changes it makes when saving a tiff, jpeg, etc.

All of these raw converter programs temporarily render the raw mage into an rgb image so we can work on them, but those changes are not baked in and stay fluid until a new file type is created with export or save as.
Click to expand...
I’m wondering what the Nikon NX Save and Save As is doing.
 
JoelKlein said:
I’m wondering what the Nikon NX Save and Save As is doing.
Click to expand...

It is saving the demosaiced image using whatever RGB file format you pick, jpeg or tiff, etc. It has to be demosaiced because each pixel has to have a red and green and blue component. It is the raw converter (nx studio or lightroom or any raw converter) that takes the monochrome pixels of the raw file and converts them into pixels with all 3 colors.
 
Member's gallery
bleirer said:
It is saving the demosaiced image using whatever RGB file format you pick, jpeg or tiff, etc. It has to be demosaiced because each pixel has to have a red and green and blue component. It is the raw converter (nx studio or lightroom or any raw converter) that takes the monochrome pixels of the raw file and converts them into pixels with all 3 colors.
Click to expand...
It’s staying NEF after using the save as.
 
JoelKlein said:
It’s staying NEF after using the save as.
Click to expand...
I see, in that case it seems it is still a raw file just with a different file name or a change to where the edits are stored, since nef is the suffix for raw files. Might be worth checking the manual. Nothing should be baked in until you output an rgb file type like tiff or jpeg. If you can still change things like the picture control profile that usually indicates it is still a raw file. Perhaps it is allowing you to save the edits in a sidecar vs. In the raw file, but the changes would not be baked in. Probably export or export as would be the way to get a jpeg or tiff that would have the edits and the demosaicing baked in.
 
Last edited:
Member's gallery
bleirer said:
I see, in that case it seems it is still a raw file just with a different file name or a change to where the edits are stored, since nef is the suffix for raw files. Might be worth checking the manual. Nothing should be baked in until you output an rgb file type like tiff or jpeg. If you can still change things like the picture control profile that usually indicates it is still a raw file. Perhaps it is allowing you to save the edits in a sidecar vs. In the raw file, but the changes would not be baked in. Probably export or export as would be the way to get a jpeg or tiff that would have the edits and the demosaicing baked in.
Click to expand...
Thats is where I am confused.
Edits are saved to the side file on its own, no need to hit save.
 
You must log in or register to reply here.
Back
Top