Demosaicing comparison Lr, DxO, C1, RT (and stuff)

[Truthiness]

I'm trying to get rid of the Lightroom subscription model, so I wanted to compare some of the higher end raw processing tools.

Unfortunately I will still remain a loyal slave to Adobe as the competition (also) suffers from issues.

DxO Photolab has problematic demosaicing - plenty of false color artifacts and also lateral CA correction sometimes causes color errors for some objects. The user interface is a bit of a mixed bag and organizing images seemed to be painful and lacking.

Capture One is monstrously expensive, I like the user interface and image quality (with reservations), but unfortunately it has some bugs - sometimes the HDR highlight-parameter has no effect on the output file and occasionally the JPG just garbage. Maybe I'll buy v21 if there's a good discount.

RawTherapee can produce splendid output, and is a life saver when one has to correct the auto focus point pattern issue from some raws. It's also hideous when it comes to usability, so it's not really an option for me.

Lightroom itself is nowdays quite sluggish and the output is pretty average and the user interface isn't quite ideal. Though, it's like an old shoe I'm accustomed to and find hard to get rid of.

Anyhow, here's the demosaicing comparison; please view at full size. The left hand side is 100%, the right hand side 200% (nearest neighbour). All sharpening and noise reduction was turned off (as far as I am aware of). RawTherapee used Amaze with false color supression level 3.

 

[Zeee]

I'm trying to get rid of the Lightroom subscription model, so I wanted to compare some of the higher end raw processing tools.

Unfortunately I will still remain a loyal slave to Adobe as the competition (also) suffers from issues.

DxO Photolab has problematic demosaicing - plenty of false color artifacts and also lateral CA correction sometimes causes color errors for some objects. The user interface is a bit of a mixed bag and organizing images seemed to be painful and lacking.

Capture One is monstrously expensive, I like the user interface and image quality (with reservations), but unfortunately it has some bugs - sometimes the HDR highlight-parameter has no effect on the output file and occasionally the JPG just garbage. Maybe I'll buy v21 if there's a good discount.

I tried it twice. For that price it doesn't list any of my telephoto lenses for lens corrections. If I was a portrait photographer I would look at it more seriously.

RawTherapee can produce splendid output, and is a life saver when one has to correct the auto focus point pattern issue from some raws. It's also hideous when it comes to usability, so it's not really an option for me.

Lightroom itself is nowdays quite sluggish and the output is pretty average and the user interface isn't quite ideal. Though, it's like an old shoe I'm accustomed to and find hard to get rid of.

Why is the user interface not ideal? You can move panels around to taste. The main reason I stick with Lr is the export page/process which is excellent. The detail panel and export page are tuned for printing.

For example DXO only has bicubic and bicubic sharper for export options. Downsampling using sharper creates a crunchy look and you to mess around with lens sharpness to send to to another editor.

I hear a lot about sluggish. It rocks on my new 2019 iMac. It even does pretty good on my travel MacBook Air that has 4GB memory. I was in a thread on another site about a PC user complaining about this. After a fresh OS install his computer woke up. There are also some basic set up, usage and simple maintenance that makes a big difference.

Anyhow, here's the demosaicing comparison; please view at full size. The left hand side is 100%, the right hand side 200% (nearest neighbour). All sharpening and noise reduction was turned off (as far as I am aware of). RawTherapee used Amaze with false color supression level 3.

--
I feel even more confident that soon things will have a wonderful conclusion

 

[knickerhawk]

At least with respect to managing false colors, the two winners here are clearly RawTherapee and LR+Enhance Details. DxO appears to trail the pack with this one image. I have and use RT on occasion when the image requires it, but like you I really dislike the interface and find it to be too much of a battle for most of my needs. Now that Adobe has added the Enhance Details option, I rarely find the need to resort to RawTherapee. Enhance Details does a remarkable job of controlling moire and other (mainly color related) aliasing issues, but it imposes a pretty significant processing time penalty and isn't practical for volume conversions.

 

[fPrime]

Thanks for posting the comparison. Lightroom + Enhance Details seems to be the demosaicing winner here.

I use Lightroom's "Enhance Details" by default in batch mode on all of my RAW's these days. It definitely helps suppress demosaic-induced color fringing for me.

fPrime

--
Half of my heart is a shotgun wedding to a bride with a paper ring,
And half of my heart is the part of a man who's never truly loved anything.

 

[His Dudeness]

I'm trying to get rid of the Lightroom subscription model, so I wanted to compare some of the higher end raw processing tools.

Unfortunately I will still remain a loyal slave to Adobe as the competition (also) suffers from issues.

DxO Photolab has problematic demosaicing - plenty of false color artifacts and also lateral CA correction sometimes causes color errors for some objects. The user interface is a bit of a mixed bag and organizing images seemed to be painful and lacking.

Capture One is monstrously expensive, I like the user interface and image quality (with reservations), but unfortunately it has some bugs - sometimes the HDR highlight-parameter has no effect on the output file and occasionally the JPG just garbage. Maybe I'll buy v21 if there's a good discount.

RawTherapee can produce splendid output, and is a life saver when one has to correct the auto focus point pattern issue from some raws. It's also hideous when it comes to usability, so it's not really an option for me.

I don't see usability issues. Anyway, you may want to take a look at ART:

Bitbucket

bitbucket.org

Which algorithm did you choose for demosaicing? There is a lot of choice in RT or ART.

BTW: darktable is another excellent open source raw converter. (Make sure to set demosaicing to Amaze, in case you try it)

Lightroom itself is nowdays quite sluggish and the output is pretty average and the user interface isn't quite ideal. Though, it's like an old shoe I'm accustomed to and find hard to get rid of.

Anyhow, here's the demosaicing comparison; please view at full size. The left hand side is 100%, the right hand side 200% (nearest neighbour). All sharpening and noise reduction was turned off (as far as I am aware of). RawTherapee used Amaze with false color supression level 3.

 

[Zeee]

Thanks for posting the comparison. Lightroom + Enhance Details seems to be the demosaicing winner here.

I use Lightroom's "Enhance Details" by default in batch mode on all of my RAW's these days. It definitely helps suppress demosaic-induced color fringing for me.

fPrime

I never really explored that thinking it was more suited for X Trans files. It looks pretty good. Thanks.

 

[fPrime]

Thanks for posting the comparison. Lightroom + Enhance Details seems to be the demosaicing winner here.

I use Lightroom's "Enhance Details" by default in batch mode on all of my RAW's these days. It definitely helps suppress demosaic-induced color fringing for me.

fPrime

I never really explored that thinking it was more suited for X Trans files. It looks pretty good. Thanks.

It does add a little bit of processing time (10 seconds per low res RAW or 20 seconds per medium res RAW) but done in batch that’s fairly painless.

fPrime

 

[Zeee]

Thanks for posting the comparison. Lightroom + Enhance Details seems to be the demosaicing winner here.

I use Lightroom's "Enhance Details" by default in batch mode on all of my RAW's these days. It definitely helps suppress demosaic-induced color fringing for me.

fPrime

I never really explored that thinking it was more suited for X Trans files. It looks pretty good. Thanks.

It does add a little bit of processing time (10 seconds per low res RAW or 20 seconds per medium res RAW) but done in batch that’s fairly painless.

fPrime

It took about 5 on my new iMac. I'm only going to use it for my hobby images I print or post online at a few sites I frequent. I only post a few from each shoot.

 

[Egregius V]

I'm a loyal DxO customer and agree that PhotoLab's demosaicing needs an upgrade! It's number one on my wishlist, even though only a handful of my images are adversely affected and thus need to be processed in something else like darktable (which supports AmaZe). In the mean time, have you tried turning on the Moire removal tool? It can handle some of the false color problems. I would also enable Lens sharpening if you have a matching optics module for your camera and lens. (This tool does more than you might realize.) And maybe experiment with the Chromatic aberration removal tool a bit while zoomed in to about 100%.

PL's primitive asset management is also slowly being worked on with a lot of feedback from customers. If that's something you need, Adobe and others provide free or low-cost solutions that can be used in tandem with PhotoLab.

As for the PhotoLab tool layout, do you know that can be customized to your liking, more or less?

--
On Flickr: https://www.flickr.com/photos/revgv/

 

[timbeach]

Delete this, I replied in the wrong place.

 

[timbeach]

The two things that seem to define the 'look' of each Raw convertor (before making your own adjustments) are the demosaicing algorithm and the embedded tonal curve. I don't think you can make proper comparisons without noting which algorithm was used, or which profile (tonal curve). This type of comparison is fine for programs which don't give you a choice (of which there are many), but not so much for programs that do (of which there are a few, such as RawTherapee). In programs with choice, you can get vastly different results within the one program.

My preferred algorithms are LMMSE and Amaze, and my preferred tonal curve is neutral, both of which RawTherapee has, which is why it is my primary option. (Ditto lens profiles, also very important). Not to mention its tremendous amount of other features, all for free. It did take a while to get used to the interface, but now its a breeze. There's a learning curve with any new program.

 

[MauriceRR]

For a faire comparison, you also need to evaluate on other subject and pictures (buildings, portraits, etc) and see how is managed texture/noise trade-off, artifacts like aliasing, etc.

 

[sluggy_warrior]

I'm trying to get rid of the Lightroom subscription model, so I wanted to compare some of the higher end raw processing tools.

as His Dudeness mentioned, you're missing the big elephant in the room: darktable. It's perhaps the most active project among free/open-source photography tools.

A Comprehensive Intro to Darktable: A Free Lightroom Alternative

Anthony Morganti of IAmMrPhotographer.com recently teamed up with photographer and fellow YouTuber Rico Richardson to produce a comprehensive introduction

petapixel.com

 

[timbeach]

I love darktable as well. I'm pretty new to it, but the more I play, the more I seem to get best results there.

 

[dRomano]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

 

[knickerhawk]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

It's impossible with converters that only offer you one demosaicing algorithm to differentiate the impact the demosaic algorithm has on micro-detail and noise (color or luminance) from the impact of post-demsoaicing operations on those same IQ attributes. It makes no practical difference anyway because you can't separate them for adjustment purposes. Really, the only useful things to look for that clearly relate to the demosaicing algorithm is color artifacts in fine high-contrast non-patterned detail (such as the crossing leaves and tree branches in the OP's image), color moire in repetitive pattern detail (such as feathers, fabric and buildings) and aliasing (such as diagonal powerlines).

 

[dRomano]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

It's impossible with converters that only offer you one demosaicing algorithm to differentiate the impact the demosaic algorithm has on micro-detail and noise (color or luminance) from the impact of post-demsoaicing operations on those same IQ attributes. It makes no practical difference anyway because you can't separate them for adjustment purposes. Really, the only useful things to look for that clearly relate to the demosaicing algorithm is color artifacts in fine high-contrast non-patterned detail (such as the crossing leaves and tree branches in the OP's image), color moire in repetitive pattern detail (such as feathers, fabric and buildings) and aliasing (such as diagonal powerlines).

Well, with some inconvenience, I did separate the rest of the processing from the initial demosaicing. That is, I exported the DXO file as a DNG so I could compare it (to the version of the raw file) in Capture One. One would not normally do that, but it's possible. I have, on rare occasion, used DXO for noise reduction or geometry corrections, or even DCP color profile, then exported the DNG to Capture One for the rest of the editing. BTW, doing this can interfere with the ability to correct chromatic aberrations and I've found that if DXO fails to correct the abberations, then the ability to fix them later is diminished. Maybe something to do with the file becoming a DNG, I don't know.

The rest of the image editing process is much more important, so any minor differences in the demosaicing doesn't really matter. However, I have on a number of occasions noticed DXO files to be more crunchy, and with artifacts after processing with PRIME noise reduction. In the past I was a firm believer of PRIME, I thought it was great and by far the best. Now I'm not so sure, since Capture One improved their algorithm. I find Capture One noise reduction to be more natural. DXO tends to produce artifacts and some of the clumps that I mentioned before still remain. Maybe PRIME thinks those clumps produced by the demosaicing are actually legit detail? As I said, I do still use it sometimes when it suits the image.

 

[IanYorke]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

It's impossible with converters that only offer you one demosaicing algorithm to differentiate the impact the demosaic algorithm has on micro-detail and noise (color or luminance) from the impact of post-demsoaicing operations on those same IQ attributes. It makes no practical difference anyway because you can't separate them for adjustment purposes. Really, the only useful things to look for that clearly relate to the demosaicing algorithm is color artifacts in fine high-contrast non-patterned detail (such as the crossing leaves and tree branches in the OP's image), color moire in repetitive pattern detail (such as feathers, fabric and buildings) and aliasing (such as diagonal powerlines).

Well, with some inconvenience, I did separate the rest of the processing from the initial demosaicing. That is, I exported the DXO file as a DNG so I could compare it (to the version of the raw file) in Capture One. One would not normally do that, but it's possible. I have, on rare occasion, used DXO for noise reduction or geometry corrections, or even DCP color profile, then exported the DNG to Capture One for the rest of the editing. BTW, doing this can interfere with the ability to correct chromatic aberrations and I've found that if DXO fails to correct the abberations, then the ability to fix them later is diminished. Maybe something to do with the file becoming a DNG, I don't know.

The rest of the image editing process is much more important, so any minor differences in the demosaicing doesn't really matter. However, I have on a number of occasions noticed DXO files to be more crunchy, and with artifacts after processing with PRIME noise reduction. In the past I was a firm believer of PRIME, I thought it was great and by far the best. Now I'm not so sure, since Capture One improved their algorithm. I find Capture One noise reduction to be more natural. DXO tends to produce artifacts and some of the clumps that I mentioned before still remain. Maybe PRIME thinks those clumps produced by the demosaicing are actually legit detail? As I said, I do still use it sometimes when it suits the image.

As PRIME works on the raw data, rather than the demosaiced image that all other converters use, you are comparing DXO's standard HQ noise reduction demosaiced image. PRIME working on the raw data is the reason for its noise reduction performance as it enhances it's ability to differentiate noise and detail.

My primary raw converter for the last few years has been C1. C1 20 and its noise reduction is significantly improved over previous versions and C1 is an excellent image editor but I find it can't match PRIME on very noisy images.

Ian

 

[dRomano]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

It's impossible with converters that only offer you one demosaicing algorithm to differentiate the impact the demosaic algorithm has on micro-detail and noise (color or luminance) from the impact of post-demsoaicing operations on those same IQ attributes. It makes no practical difference anyway because you can't separate them for adjustment purposes. Really, the only useful things to look for that clearly relate to the demosaicing algorithm is color artifacts in fine high-contrast non-patterned detail (such as the crossing leaves and tree branches in the OP's image), color moire in repetitive pattern detail (such as feathers, fabric and buildings) and aliasing (such as diagonal powerlines).

Well, with some inconvenience, I did separate the rest of the processing from the initial demosaicing. That is, I exported the DXO file as a DNG so I could compare it (to the version of the raw file) in Capture One. One would not normally do that, but it's possible. I have, on rare occasion, used DXO for noise reduction or geometry corrections, or even DCP color profile, then exported the DNG to Capture One for the rest of the editing. BTW, doing this can interfere with the ability to correct chromatic aberrations and I've found that if DXO fails to correct the abberations, then the ability to fix them later is diminished. Maybe something to do with the file becoming a DNG, I don't know.

The rest of the image editing process is much more important, so any minor differences in the demosaicing doesn't really matter. However, I have on a number of occasions noticed DXO files to be more crunchy, and with artifacts after processing with PRIME noise reduction. In the past I was a firm believer of PRIME, I thought it was great and by far the best. Now I'm not so sure, since Capture One improved their algorithm. I find Capture One noise reduction to be more natural. DXO tends to produce artifacts and some of the clumps that I mentioned before still remain. Maybe PRIME thinks those clumps produced by the demosaicing are actually legit detail? As I said, I do still use it sometimes when it suits the image.

As PRIME works on the raw data, rather than the demosaiced image that all other converters use, you are comparing DXO's standard HQ noise reduction demosaiced image. PRIME working on the raw data is the reason for its noise reduction performance as it enhances it's ability to differentiate noise and detail.

My primary raw converter for the last few years has been C1. C1 20 and its noise reduction is significantly improved over previous versions and C1 is an excellent image editor but I find it can't match PRIME on very noisy images.

Ian

Correct, PRIME works on the raw image, I wasn't stating otherwise. Since DXO can't import a DNG, there would be no way to run PRIME on any image that is not a raw file.

The DNG that I output from DXO to compare with C1, had no noise reduction enabled in DXO.

 

[knickerhawk]

I've tried to make comparisons of the demosaicing of LR, DXO and C1 before and found there is not much difference. It's important to mention that I was only looking at what I perceived as sharpness, detail retention and noise. I did not consider color, which is an unavoidable part of the process.

All types of corrections were OFF. This is very important, because I was trying to isolate the demosaicing as the only variable. I'm not 100% sure I achieved that, but I did all I could. The only parameter that I could not set to OFF was color profile, so used whatever the standard profile was.

Mainly what I found was that DXO produced images that at first seemed sharper. They had a feeling of higher accutance, but it was really because they were crunchier, not because there was more detail. The noise in the out of focus areas tended to be crunchier, with larger clumps of grain. The overall color noise (subjectively) was about the same as Capture One. My preference is Capture One because the noise is "gentler". I'm not saying it has "less" noise (I did not quantify it with standard deviation measurements) but the noise was more uniform and didn't have the clumps that the DXO PL3 had. I should note that Capture One 20 was used, and I did not compare with previous versions, but I have tested the noise reduction and it does render nicer, smoother noise with better detail retention than older versions of Capture One.

Obviously there are other important factors besides demosaicing. I've only tested that issue alone.

It's impossible with converters that only offer you one demosaicing algorithm to differentiate the impact the demosaic algorithm has on micro-detail and noise (color or luminance) from the impact of post-demsoaicing operations on those same IQ attributes. It makes no practical difference anyway because you can't separate them for adjustment purposes. Really, the only useful things to look for that clearly relate to the demosaicing algorithm is color artifacts in fine high-contrast non-patterned detail (such as the crossing leaves and tree branches in the OP's image), color moire in repetitive pattern detail (such as feathers, fabric and buildings) and aliasing (such as diagonal powerlines).

Well, with some inconvenience, I did separate the rest of the processing from the initial demosaicing. That is, I exported the DXO file as a DNG so I could compare it (to the version of the raw file) in Capture One. One would not normally do that, but it's possible. I have, on rare occasion, used DXO for noise reduction or geometry corrections, or even DCP color profile, then exported the DNG to Capture One for the rest of the editing. BTW, doing this can interfere with the ability to correct chromatic aberrations and I've found that if DXO fails to correct the abberations, then the ability to fix them later is diminished. Maybe something to do with the file becoming a DNG, I don't know.

The rest of the image editing process is much more important, so any minor differences in the demosaicing doesn't really matter. However, I have on a number of occasions noticed DXO files to be more crunchy, and with artifacts after processing with PRIME noise reduction. In the past I was a firm believer of PRIME, I thought it was great and by far the best. Now I'm not so sure, since Capture One improved their algorithm. I find Capture One noise reduction to be more natural. DXO tends to produce artifacts and some of the clumps that I mentioned before still remain. Maybe PRIME thinks those clumps produced by the demosaicing are actually legit detail? As I said, I do still use it sometimes when it suits the image.

As PRIME works on the raw data, rather than the demosaiced image that all other converters use, you are comparing DXO's standard HQ noise reduction demosaiced image. PRIME working on the raw data is the reason for its noise reduction performance as it enhances it's ability to differentiate noise and detail.

My understanding is that DxO Prime is not operating before the demosaicing stage. Like every other NR tool I know of that's incorporated into a raw converter, it does its work after the demosaicing stage. When you output a DNG from DxO, it's a linear DNG, which means it has already been demosaiced but gamma correction and white balancing hasn't been applied yet.

My primary raw converter for the last few years has been C1. C1 20 and its noise reduction is significantly improved over previous versions and C1 is an excellent image editor but I find it can't match PRIME on very noisy images.

Ian