Canon EOS R3 Baked in Raw Noise Reduction Revisited

Thanks for showing this!

This shows clearly the difference between a non-baked and a baked raw file!

The non-flat Fourier Power Spectra tells us that something is going on, i.e that there is some correlation between the pixels.

As a side note, for the R6, the 9 pixels taking part in the red or blue averaging kernels are well separated by one pixel, whereas the 9 pixels taking part in the green averaging kernel are touching other: https://www.dpreview.com/forums/post/65857338

I think this makes the Fourier transform complicated to interpret, especially if pixels from all 3 colors are taken as a whole, and I suspect this could be one of the reasons why you have a difference between sensor rows and sensor columns in your FTs for the R3.

Also, the auto-correlation method is a great tool to complement the Fourier Transform, as ideally it can provide an estimate for the averaging kernel.

Once NR is detected, the tricky part in NR analysis is to isolate the specific color channels and/or specific rows/columns on which raw processing is applied. It’s a a bit of a headscratching and trial-and-error exercise…

Ethan

FWIW, Canon EOS R6

Note vertical transitions from "valley" to "mustache"

These analysis combine all channels so that obscures some of the PDAF stuff you refer to.

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Yeah, since it can't be turned off there is no objective way to measure how detail is affected.

It's certainly true that there are no wide-spread complaints about lost detail so that's a good sign.

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Engineers fine tune the output from every camera. The goal is to produce good looking and easy to post process images. Guess we do notice even when we are not aware of noticing.

Completely uncooked RAW files are ugly and need a lot of additional post processing. Some who want uncocked RAW do probably not quite know what they wish for.

Tristimulus wrote:

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Engineers fine tune the output from every camera. The goal is to produce good looking and easy to post process images. Guess we do notice even when we are not aware of noticing.

Completely uncooked RAW files are ugly and need a lot of additional post processing. Some who want uncooked RAW do probably not quite know what they wish for.

FWIW, I have tested over 400 camera models and "cooked" raw files are the exception not the rule. The vast majority of the raw files I encounter have no detectable signal processing.

Lost99999 wrote:
....

Do you think things are ‘lost’ or ‘worse’ for the average and expert user when Canon ‘bakes’ the Raw file ?

Mathematically, detail must be lost due to the signal processing but it's not possible to quantify (although I have an idea of how this might be done).

So, it's a case of not knowing what you are missing.
Certainly it's not so much that people are complaining.

Technically it does make comparisons to other sensors "apples to oranges".

bclaff wrote:

Tristimulus wrote:

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Engineers fine tune the output from every camera. The goal is to produce good looking and easy to post process images. Guess we do notice even when we are not aware of noticing.

Completely uncooked RAW files are ugly and need a lot of additional post processing. Some who want uncooked RAW do probably not quite know what they wish for.

FWIW, I have tested over 400 camera models and "cooked" raw files are the exception not the rule. The vast majority of the raw files I encounter have no detectable signal processing.

Have had some astronomical CCD cameras where differences in photosite sensitivity, colomn errors, hot and dark pixels were part of the RAW files. Had to be corrected. Not so with any of the ordinary stock cameras with CCD sensors I did use.

RAW files from CMOS astro cameras look more like ordinary stock cameras. Even onion rings, polygons and traces of scaling and shading correction seems to be baked in the RAW files. The cameras are low end astro cameras as the real stuff cost $10 000 ++ apiece.

But you are the expert here, I am not.

Just wondering - how do you know that there is no on chip processing going on?

bclaff wrote:

It's certainly true that there are no wide-spread complaints about lost detail so that's a good sign.

indeed - not even with Pentax K-3M3 which is the current reigning champion of RAW-NR

https://www.photonstophotos.net/Charts/PDR.htm#Canon%20EOS%20R3,Pentax%20K-3%20III

Tristimulus wrote:

bclaff wrote:

Tristimulus wrote:

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Engineers fine tune the output from every camera. The goal is to produce good looking and easy to post process images. Guess we do notice even when we are not aware of noticing.

Completely uncooked RAW files are ugly and need a lot of additional post processing. Some who want uncooked RAW do probably not quite know what they wish for.

FWIW, I have tested over 400 camera models and "cooked" raw files are the exception not the rule. The vast majority of the raw files I encounter have no detectable signal processing.

Have had some astronomical CCD cameras where differences in photosite sensitivity, colomn errors, hot and dark pixels were part of the RAW files. Had to be corrected. Not so with any of the ordinary stock cameras with CCD sensors I did use.

RAW files from CMOS astro cameras look more like ordinary stock cameras. Even onion rings, polygons and traces of scaling and shading correction seems to be baked in the RAW files. The cameras are low end astro cameras as the real stuff cost $10 000 ++ apiece.

But you are the expert here, I am not.

Just wondering - how do you know that there is no on chip processing going on?

"Hot" pixel mapping etc. certainly  happens.
Normally lens distortion correction etc is deferred.

In any case, the signal processing that matters in my testing is that which is applied to all pixels (as opposed to "hot" pixel mapping for example).

Pillar Lee wrote:

Thanks for your work, bill, but I wonder where I could find the 3D chart energy spectra of Lumix S1R. I searched photonstophotos but got nothing.

They aren't available online but here they are:

Notice how the signal processing is only at low ISO settings (through ISO 200)

You can see 2D base ISO only at Energy Spectrum

bclaff wrote:

FWIW, Canon EOS R6

Thanks!

Note vertical transitions from "valley" to "mustache"

These analysis combine all channels so that obscures some of the PDAF stuff you refer to.

Okay, thx for clarifying.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

bclaff wrote:

Tristimulus wrote:

bclaff wrote:

Tristimulus wrote:

KZ7 wrote:

These findings are interesting from a technical perspective.

The main questions I have is how much difference does this make to the level of detail in the raw files compared with for instance the Leica and what are the benefits / drawbacks of doing this?

In real world use, will we even notice?

Engineers fine tune the output from every camera. The goal is to produce good looking and easy to post process images. Guess we do notice even when we are not aware of noticing.

Completely uncooked RAW files are ugly and need a lot of additional post processing. Some who want uncooked RAW do probably not quite know what they wish for.

FWIW, I have tested over 400 camera models and "cooked" raw files are the exception not the rule. The vast majority of the raw files I encounter have no detectable signal processing.

Have had some astronomical CCD cameras where differences in photosite sensitivity, colomn errors, hot and dark pixels were part of the RAW files. Had to be corrected. Not so with any of the ordinary stock cameras with CCD sensors I did use.

RAW files from CMOS astro cameras look more like ordinary stock cameras. Even onion rings, polygons and traces of scaling and shading correction seems to be baked in the RAW files. The cameras are low end astro cameras as the real stuff cost $10 000 ++ apiece.

But you are the expert here, I am not.

Just wondering - how do you know that there is no on chip processing going on?

"Hot" pixel mapping etc. certainly happens.
Normally lens distortion correction etc is deferred.

In any case, the signal processing that matters in my testing is that which is applied to all pixels (as opposed to "hot" pixel mapping for example).

Thank you for responding. Illuminating answer.

You do great service to the photo community - no offence intended.

Keep up the good work!

The Leica curves are on 0.6 and 0.8 vertical scales, while the R3 ones are on 0.35 and 0.3, which visually stretches them vertically twice or so, exaggerating the non-flatness.