LR vs PureRAW 3 noise reduction test

[BeatX]

Hi,

Another test, this time I wanted to check which AI noise reduction tool is better between most popular and most rated software on the market right now: Lightroom vs DXO PureRAW 3

Since literally 99% of denoising test results are shown using close-ups of birds in natural (but dim) lighting, I suspect that these types of photos are the most "convenient" or "optimal" to present the results.
So, to put the denoising software to maximum stress test, I chose a photo that is one of the most challenging to denoise: a severely underexposed photo in crazy, artificial stage lighting where the shot is full of distant details, like human faces, t-shirt patterns, etc.
Link to .RAF file:

https://drive.google.com/file/d/1wlKmbPpwZ7fw55xWuPmwrUL-TD3POQ8s/view?usp=sharing

All presented files are exported to .jpg from Lightroom, with sharpening set to 0 and no lens corrections applied

Original file:



Original file pushed +3 EV



Lightroom + enhance tool pushed +3 EV



DXO PureRAW 3 (DeepPRIME XD) pushed +3 EV



For those of You, who don't want to pixel peep


Original file pushed +3 EV


LR + Enhance tool pushed +3EV


DXO PureRAW 3 (DeepPRIME XD) pushed +3 EV

For me the conclusions are obvious: LR wins hands down
The amount of errors in the denoising process in DXO PureRAW 3 leaves a lot of strange and unwanted artifacts in the photo, which does not look good.
Plus, Lightroom produces slightly more overall fine details.
In addition, DXO PureRAW 3 changes the colors of photos quite a bit, which I'm not a fan off, and which I noticed some time ago, that .dng files processed by DXO PureRAW 3 have some color shift compared to the original.

Hope it helps

--
X-H2 | Viltrox 13/1.4 | Viltrox 27/1.2 | Viltrox 75/1.2
My gallery: https://www.flickr.com/photos/maciej_k/

 

[RWN Photo]

LR for the win... this time. I'm sure a different image, will result in different results and software bashing. None of the post processing software is THE best.

 

[Morris0]

Three stop push. No!

Morris

 

[BeatX]

Three stop push. No!

Morris

Why not?

 

[Migralda]

Have you tried Topaz photo AI ? I prefer it to DXO it has a lot more individual adjustment options.

 

[Rightsaidfred]

Thank you Beat for this exercise.

I've also shot a rock band multiple times, and I know the quirks.

Former darktable user, went for LR end of 2023, one reason being that I wanted to have the AI denoise function. Very happy, too.

Cheers,

Martin

 

[BeatX]

Have you tried Topaz photo AI ? I prefer it to DXO it has a lot more individual adjustment options.

No, I haven't

 

[BeatX]

Thank you Beat for this exercise.

I've also shot a rock band multiple times, and I know the quirks.

Former darktable user, went for LR end of 2023, one reason being that I wanted to have the AI denoise function. Very happy, too.

Cheers,

Martin

Sure thing Martin

 

[Mark Casebeer]

I find both software has its advantages and disadvantages. For me alot depends on the ISO levels. I find LR does the best on ISO levels 4000 and above. PureRaw 3 does a better job at lower levels with its sharpening applied. LR lets you crop and apply adjustments before Denoise which I like. I agree with you that PureRaw does come out of the process affecting the colors and it comes from the original unprocessed state.

Mark

 

[BeatX]

Ok, so I've prepared another pack of sample images.
This time however, they are properly exposed, and actually good enough to publish (not like my example from first post, where I would delete this image right after taking a shot, but I do decided to keep it to examine denoising capabilities).
All files are exported in LR with sharpening set to 0, and all lens profile corrections are disabled:


Original file


LR + Enhance


DXO PureRAW 3 (Deep Prime XD)

Please, pay attention to texture of my wife green coat.
I can see clearly more details, more definition from LR version.
Not much, but still.


Original file


LR + Enhance


DXO PureRAW 3 (DeepPRIME XD)

Same story, fabric texture has slightly more details from LR version.


Original file


LR + Enhance


DXO PureRAW 3 (DeepPRIME XD)

And again, same thing.
LR version has more fine details (it can be clearly seen on my dogs fur), plus I cannot spot any artifacts, where from PureRAW 3 version I can see some unwanted artifacts here and there.

So Morris, as you can see - my first photo example may be exaggerated for normal denoising usage - because that particular photo is not suitable for anything other than testing denoising software.
But this photo is excellent material for testing denoising software, because differences in denoising this heavy underexposed and detail rich photo, are much more pronounced, which makes it easier to give a verdict for everyone

LR is just better than DXO PureRAW 3. Period.
Not sure how Topaz Denoise AI would handle.
Someone could do similar test to mine, and show us the results..

One more thing: apologies for huge color differences in my first image.
I've just realized, that LR is using by default "Adobe color" profile for .dng files, instead the one which was set in my camera during taking the shot.
So in practice I was watching my images in LR with for example "Provia" film sim in original .raf file and with "Adobe color" for DXO PureRAW processed .dng file.
Lightroom "Enhance" tool, is using by default same color profile as was selected in camera during taking the shot.

So guys, forget what I wrote in the first post, that the DXO version is different in color from LR version.
My bad
Although, I don't know why - but DXO version has always slightly changed WB value than original file, and I have to correct it every time.

--
X-H2 | Viltrox 13/1.4 | Viltrox 27/1.2 | Viltrox 75/1.2
My gallery: https://www.flickr.com/photos/maciej_k/

 

[Truman Prevatt]

The measure of fidelity in the audio industry is given by: Audio fidelity is the measure of how accurately a recording reproduces the original sound source. This is the quality of sound music reproduction. That's pretty much the same in the RF industry. When FCI was developing cameras and sensors back in the 50's through the 90's, the fidelity was evaluated based on a ground truth test range we hadn't on on the end of Long Island, NY. A higher fidelity indicates better quality be it audio, RF or EO imagery or radar imagery, while a lower audio fidelity indicates poorer quality.

The fidelity of how a camera produces an image needs to have the original source to compare to. While the samples we see here are interesting, the results of the conclusion are purely subjective since there is not ground truth source to compare. There is no way to tell if the "details" one sees in one and not the other are real or artifacts of the processing or artifacts in the database driving the processing. An artifact taken as detail is a decrease in fidelity not an increase although it might me more pleasing to the eye.

Too often most of the comparisons on the forums of X vs. Y boils down to I prefer X vs. Y with no fidelity metric to compare to a ground truth.

Are they useful? Sure but the caveat should remain that they boil down to a subjective preference to rendering X over Y because . It appeals is not the same as it produces the ground truth with better fidelity.

There are a few points to take into consideration. There is no information in an image (read detail if you want) above the maximal wavenumber of the physical sensor. Detail requiring wavenumber above the maximal wave number are lost forever. Reproducing detail with any prediction algorithm is only a guess. For producing an artistic print - that might be good enough. However, for producing cameras and imaging systems for reconnaissance and surveillance - artistic interpretation information above the wavenumber cutoff or from below the noise floor are not only unusable but are totally unreliable.

I had some images of three wild horsed galloping across the high desert in WY. I tried Topez and Pure Raw and neither produced more detail - they produced artifacts. In a print the image might have looked better. But in reality the new detail was no more than artifacts that looked as if they could be real. I expect the same is true for Enhanced Details. I discarded the images.

This is not to say we don't appreciate the effort that BeatX and others put in to give us comparisons. We should simply consider that detail above the information content of the sensor (including the SNR) is lost forever and conclusions may be simply a subjective - "I like it better." Of course there is nothing wrong with liking something better.

 

[Morris0]

Three stop push. No!

Morris

Why not?

Look at the results.

Morris

 

[Truman Prevatt]

I find both software has its advantages and disadvantages. For me alot depends on the ISO levels. I find LR does the best on ISO levels 4000 and above. PureRaw 3 does a better job at lower levels with its sharpening applied. LR lets you crop and apply adjustments before Denoise which I like. I agree with you that PureRaw does come out of the process affecting the colors and it comes from the original unprocessed state.

Mark

I had some shots of big horn sheep walking along grazing. I had thought I had changed the shooting bank to insure a higher ISO hence a higher shutter speed. It turns out I hadn’t and there was some slight motion blur. Pure Raw did nothing. However, Topez Photo AI’s motion blur setting cleaned the small amount of motion blur nicely. All these applications do some things better than the others.

 

[BeatX]

Three stop push. No!

Morris

Why not?

Look at the results.

Morris

https://www.dpreview.com/forums/post/67874951

So Morris, as you can see - my first photo example may be exaggerated for normal denoising usage - because that particular photo is not suitable for anything other than testing denoising software.
But this photo is excellent material for testing denoising software, because differences in denoising this heavy underexposed and detail rich photo, are much more pronounced, which makes it easier to give a verdict for everyone

 

[BeatX]

All these applications do some things better than the others.

This is very true, and that is why Im using both DXO PureRAW 3 and LR simultaneously.
For low ISO images when I want to squeeze as much as I can from the sensor - to have all those nice fine details in landscape photos - I find PureRAW 3 being slightly better than LR, due to excellent DXO optical module lens corrections + sharpening
(check results from my previous test: https://www.dpreview.com/forums/post/67858047 )
But, for high ISO images I choose LR denoising tool, to have slightly more fine details and less unwanted artifacts.
Although since last year I was relaying for demosaicing and denoising only on DXO PureRAW 3, and I was very happy with results (many o them I've published in here), so it's not that DXO software now become worst for me.
It is rather that I have now slightly better denoising tool, due to my switch from C1 to LR.
DXO PureRAW 3 is still superb RAW processing software and I do highly recommend every Fuji X user, to buy it and use it with whatever RAW developer.

Anyway Imho this workflow (LR + DXO PureRAW3) combined with premium lenses available for Fuji X system, gives almost identical (if not identical) results in low ISO images as hi-res FF sensor + best FF lenses, and very close to FF setup in high ISO images.
FF system has an obvious advantage in shallower DOF (it is more pronounced with fast, wide angle primes), but with lenses like Viltrox 27/1.2 and Viltrox 75/1.2 gap is getting smaller.
At least in standard and short telephoto range.

Im curious, if Viltrox will design lens like 16/1.0 which would make a lot of people veeery happy (me included)

--
X-H2 | Viltrox 13/1.4 | Viltrox 27/1.2 | Viltrox 75/1.2
My gallery: https://www.flickr.com/photos/maciej_k/

 

[baobob]

In my experience to avoid artefacts in DPR3 I use the following settings





and I finish the PP in PS

I get similar results with both solutions DXo or Adobe AI NR but I mainly use DXO PR3 which is better integrated to my WFlow.

Sharpening in PR3 does create artefacts

--
Good judgment comes from experience
Experience comes from bad judgment

Bob RJF

www.flickr.com

 

[baobob]

With extreme dim light, Iredid some trials and I found the same conclusions than you.here is the original in JPEG



On the left you can hardly see the clock on the roof

Now Adob AINr vs DPR3 :



Left Adobe right DXO..

Adobe had corrected the RAF (Tamron 18-300mm at 18mm) this accounts for the position on the roof

Adobe wins without any contestation. I was surprised by the difference and redone the DXO development 2 times.

In extreme situations Adobe recovers much better the image, definitly...

--
Good judgment comes from experience
Experience comes from bad judgment
https://www.flickr.com/people/robertjf/

 

[Phil1]

I use DeepPRIME XD in DxO PL7 where one can tailor the amount of noise reduction required which depends critically on the ISO. I use settings of 5 (for Low ISOs i.e. up to 800) up to 40, or a little higher for ISOs of 6,400. Use too high a setting for a given ISO and the amount of detail goes down a little but it’s clearly noticeable. So, unless you've optimised the settings you can't expect to get the best results.


With all the AI programmes if you push the sliders too far you'll get artefacts, especially with sharpening ones, and one has to exert some restraint to avoid them.


Like Truman I also sometimes use the sharpening in LR in addition to Topaz AI sharpening to get the best results as the two types of S/W work very differently.

Hope this adds to the discussion!

Phil

 

[Truman Prevatt]

At base ISO, there is little any of the programs like PureRAW or Topaz can do but generate artifacts of details that don't exist in the image. Enhanced details in LR might be different with XTrans since Adobe's long ignoring XTrans.

None of the above, however, will produce a better image than C1, or standard Lightroom on my Z8 images at base ISO (64) exposed properly. Same with XTrans and C1 and XTransformer and standard LR.

It is only in lowlight or other marginal conditions such as some motion blur of the image that one needs PureRaw, Topez, enhanced details, etc. Trust me while I own PureRaw and Topaz Photo AI, neither is on my standard workflow. They show up when either the lighting conditions demand high ISO or I screw up maybe with shutter speed by trying to keep the ISO down, to save my sorry butt. Sometimes they do, sometimes they don't. However, I sure don't strive to depend on them. Just because they might save my sorry butt, does not mean the resulting images are all that great.

 

[BeatX]

None of the above, however, will produce a better image than C1, or standard Lightroom on my Z8 images at base ISO (64) exposed properly. Same with XTrans and C1 and XTransformer and standard LR.

Hi Truman thx for Your input.

Can You upload some Z8 raw files at base ISO (preferly landscape oriented)? so I can play with PureRAW 3 and LR with Nikon files.