DXO PureRaw4 vs Adobe Enhanced NR

[JimFox]

I was fortunate that I am starting to move my backups from California to Colorado. So I had my backup drives from the last 2 years.

I had mentioned in another thread about terraced patterns being introduced by DXO PureRaw4, so I was able to back and find the images that I had noticed them in last year. This is from PhotoLabs 7, which Alan can correct me, was supposed to use PureRaw4 as the Engine for the top NR in it. I do have the stand alone PureRaw3 and the results were the same. I am including what I ended up using, Adobe ACR image uses the Enhanced Mode.

For a point of reference the image is about 3 stops underexposed for the ground.

There was a 2nd issue I had forgotten about, the Terraced ground reminded me of the Philippines, so that's why that stuck in my mind. And I actually could have covered that up to an extent by darkening the ground more. The worst issue was actually the stars. DXO turned most of the fainter stars along the edges into what looks like Tadpoles. It adds tails to most of the stars.

When I was processing this, I had kind of covered up the terraces by taking the ground darker, and it wasn't until I was finishing up that I happened to zoom in on the stars and then see all of the stars being ruined by the made up tails. I had never had that issue before with Photoshop and ACR, so I didn't even know to check. That was a bunch of hours down the drain, because there was no correcting that. I had to go back and restart the whole process on this timelapse set of images in ACR then.

The NR was set in both to about 70%, I don't recall the exact amount. But I adjusted the slider until the Noise disappeared. The WB is a bit different between the 2, but similar. But that doesn't affect the Noise Reduction.

Captured up on Cinamon Pass in SW Colorado in July 2024
Nikon D850
Sigma 14mm f1.8
ISO 2000
20 secs
f1.8

1. DXO PureRaw Full image

2. Adobe ACR Full image

3. DXO PureRaw Lower Left - Notice the terraced patterns, and then there is even a straight line like it's a section.

4. Adobe ACR Lower Left

5. DXO PureRaw Lower Right Side Sky - Notice all of the tails coming off so many of the stars. In other areas it seemed to have even added stars and texture that wasn't there.

6. Adobe ACR Lower Right Side Sky

 

[Kyle Jones]

I'll say that I don't like the ground in either of them. I'd rather have some noise than the smeared look that is there.

I've seen the same sky behavior from pureRaw.

 

[Jameel Hyder]

Agree with Kyle. Underexposed 3 stops is pushing the limit. Its either going to get smeared or the AI will try to create something there so it is not smeared. I rarely push shadows even 2 stops.

 

[JimFox]

I'll say that I don't like the ground in either of them. I'd rather have some noise than the smeared look that is there.

I've seen the same sky behavior from pureRaw.

I agree both of these are too plasticky. If that's a word...

I personally would rather work (which I did) with a ground layer that has less details then one that has made up details, especially that terraced pattern. If I put out a photo of the mountains with what looks like rice terraces, people are going to say fake.

So I could work with ACR and make it so it didn't look as smeared by recovering the ground in steps. But the terrace pattern was there in the DXO version no matter what I did.

 

[AlanLichty]

What does original image look like down in the lower left corner? I'm with both opinions above - neither look very nice to me at all.

 

[JimFox]

Agree with Kyle. Underexposed 3 stops is pushing the limit. Its either going to get smeared or the AI will try to create something there so it is not smeared. I rarely push shadows even 2 stops.

Thanks Jameel. Read my comments to Kyle, I don't want to repeat them again.

I haven't paid much attention generally speaking as to how many stops I am underexposing the ground. So I am not sure what my limit is. I will pay more attention to that this year.

 

[JimFox]

What does original image look like down in the lower left corner? I'm with both opinions above - neither look very nice to me at all.

I agree, neither are favorable, but in the end, made up terrace patterns can't be covered up. At least ACR didn't invent shapes into an area. So for my working versions, I did ACR in smaller steps to not have the ground be plastic.

In the original, without any editing, the ground is just black. The ACR version is the most true to what the canera captured. If I just raised the exposure and ignored the noise, it looked like the ACR version as far as textures go. I can post a SOOC jpg, but it's just black to the eye on the ground.

What about how DXO ruined the stars?

 

[AlanLichty]

Would you be willing to copy the RAW file to a dropbox? PM me your email address so I can send a request.

 

[JimFox]

Would you be willing to copy the RAW file to a dropbox? PM me your email address so I can send a request.

Of course.

 

[AlanLichty]

My efforts to raise the dead were not exactly fruitful as the lower left corner didn't have much in the way of useful data to recover. This my take on his MW shot after running it through PureRAW4 and doing a lot of touchup work in LR and PS:

 

[Jameel Hyder]

For this particular image keeping the ground layer dark with just a hint of detail will probably work best.

 

[AlanLichty]

One more using PhotoLab 8 with a pass through Topa DeNoiseAI in Photoshop:

 

[AlanLichty]

For this particular image keeping the ground layer dark with just a hint of detail will probably work best.

Fully agree - I tried to match what Jim was doing with his image at the top but it does feel like too much light. I like the first of the edits I tried better than the version using Photolab 8. That one just feels like too way much light unless you had a full moon.

 

[JimFox]

One more using PhotoLab 8 with a pass through Topa DeNoiseAI in Photoshop:

View attachment 79765

Need the 100% crop of the lower left and the stars to compare to mine. I don't think anything would show up at this.

I do like the overall look of what you got in this though, but it's the 100% detail I would want to see so I could tell if when someone is playing the compiled timelapse on their 70" TV, that there are no rice terraces showing up in the ground layer.

 

[JimFox]

My efforts to raise the dead were not exactly fruitful as the lower left corner didn't have much in the way of useful data to recover. This my take on his MW shot after running it through PureRAW4 and doing a lot of touchup work in LR and PS:

View attachment 79763

And Alan, it doesn't have to have detail, it just has to not have made up detail. Adobe ACR did this without adding made up detail, And I could do it less aggressively so it doesn't look Plastic. That was just a fast run through on my part to show the ACR isn't making up detail.

 

[JimFox]

For this particular image keeping the ground layer dark with just a hint of detail will probably work best.

And that's what I did in my finished processing, I kept it darker which helped hide some of the noise as well. My whole point is that PureRaw was inventing and adding detail that didn't exist, while Adobe ACR doesn't. And I easily could have not pushed ACR to the point where it was looking plastic. My use case is probably extreme compared to most people, but it exists. And a huge portion of my photography is at night. I can't use a program that adds non existant detail to images or adds stars and details to the sky when I try to present an honest view of the night sky.

 

[AlanLichty]

I checked Photolab 7 and it is using the denoise version from PureRAW 3 so that was an earlier version than what I used above. I did quite a few edits on that file that would be hard to batch out to make a time-lapse out of to be honest. I wasn't really thinking about reproducing the scene hundreds of times over with my edits

 

[Kyle Jones]

I just tried out PureRAW 5 on one of my recent night images. I'm still struggling with it for dark landscapes. I think it smooths out the noise too much and then invents details that don't exist. Now that it has some sliders, I am hoping to do a parametric study with different settings to see how they all work. But so far, even with the luminance way down (10) and the force details off (0) the real detail is gone and fake artifacts are appearing.

 

[AlanLichty]

Interesting observations. I haven't seen the artifacts you describe but I don't much night imagery - only daytime and low light conditions.

PureRAW 5 has been doing a very good job on all images I have thrown at over the past several months during the beta testing versions with shots dating back to my old Canon 20D with excellent results (DxO's beta tests for PureRAW 5 started in late December). The only night shot I have played with using one of the PureRAW 5 beta variants was actually the one Jim posted above. I did give him a copy of the result but did not post it here due to the NDA that was in place during the beta test period.

 

[Kyle Jones]

Have you played with the sliders much? I'm wondering if you have any advice for low-light images