DxO DeepPrime XD Review – An Evolution of Noise Reduction

Tips & Techniques

Artificial intelligence, or rather machine learning, is now a permanent part of many photo editing programs. I recently tested one such machine learning program for noise reduction and sharpness, called DxO PureRaw 2. In the meantime, however, the French company DxO reworked their de-noising module called DeepPRIME (which is part of PureRaw 2) and introduced a new version called DeepPRIME XD.

The suffix XD stands for “extra detail.” This gives us a hint of what to expect from the new algorithm. But how does it perform? That’s what I wanted to cover in today’s quick review.

Robin_DeepPrime XD
NIKON D500 + 500mm f/5.6 @ 500mm, ISO 5000, 1/400, f/5.6

To start, I should say that I was hoping to see DeepPRIME XD integrated into DxO PureRaw 2 via an update. In fact, PureRaw 2 has become almost an integral part of my editing workflow. Unfortunately, it didn’t happen. So maybe in PureRaw 3? In any case, for now, it’s only found in DxO PhotoLab 6.

As such, DxO PhotoLab 6 has taken up temporary residence in my computer. Only time will tell if it will stay there permanently. I have left all its editing and cataloguing features aside and focused my attention solely on DeepPRIME XD.

Table of Contents

Processing Speed

I selected a set of eleven photographs for editing, taken mostly with Nikon D500 and Nikon D750 cameras. The lenses used were a Nikon 400mm f/2.8 G or a Nikon 200-500 f/5.6 zoom lens. I mention this fact because both the camera and the lens must be in the program’s database in order to use its full potential. That’s why I didn’t include the Nikon Z9 with the new telephoto lenses in the test. While the Nikon Z9 is familiar to DxO, the Nikon 400 f/4.5 and 400 f/2.8 TC S lenses were not yet in the database at the time of my testing.

I processed the entire series of photos twice to minimize any possible error. The resulting processing times were different for each method, and not by a small amount. DxO PhotoLab 6 processed the photos the fastest, at 1 min. 12sec., when I set the older DeepPRIME algorithm. The same algorithm, but in DxO PureRaw 2, completed the job in 1 min. 52 sec. And finally, my computer’s processor had the most work to do with the new DeepPRIME XD. It took a long 3 min. 56 sec. – so, the new XD algorithm is hardly fast.

But what matters the most is image quality. Does it fare better there?

Noise Reduction Quality

DxO claims that the standard DeepPRIME algorithm should reduce the noise level by 2 stops, while DeepPRIME XD will show a difference of 2.5 stops or more. These are marketing numbers, of course, but they at least show that DxO considers the XD version of the algorithm to be superior.

How does this difference be reflected in the photos themselves? In my opinion, the difference is noticeable immediately, if you zoom in enough.

Take a look at the 100% crops below. DeepPRIME XD (the image on the right) clearly has the edge over the older algorithm on the left.

Great Tit_PureRaw2Great Tit_DP XD

What ISO did I use for this photo? A whopping ISO 20,000! I don’t shoot at ISO 20,000 every day, but I can imagine situations where it comes in handy, especially in dense forests.

Although the DeepPRIME XD algorithm looks cleaner and less “blocky” at first glance, note that it also created some artifacts under the bird’s beak and behind its head. The older DeepPRIME algorithm may not look as clean, but the artifacts are not as obvious in the image.

Preservation of Details

The usual choice in noise reduction is pretty simple: Either you get a photo with minimal noise and minimal detail, or more detail but more noise. If you want both, you typically need to use masks and apply noise reduction only where it’s the most needed, like out-of-focus regions.

Artificial intelligence algorithms have improved this tradeoff, but it’s still an unavoidable part of photography. This is where I was most curious to see how DeepPRIME XD would compare to the previous version.

The original DxO DeepPRIME is on the left, and DxO DeepPRIME XD is on the right:

Bicolored Antbird_DeepPrimeBicolored Antbird_DeepPrime XD

When I compare the results, I’m not disappointed. You can see that the edge sharpness has increased slightly in the XD version, and the amount of noise in the out-of-focus background has decreased. Unlike the prior test, this is definitely a win for the newer algorithm.

Here’s another example below. Check out the crest of the Andean Cock-of-the-rock. I think the difference is really visible. (As before, the older algorithm is on the left, and the newer XD is on the right.)

Cock-of-the-rock DeepPrimeCock-of-the-rock DeepPrime XD

Color Preservation

This is where things get a little complicated. I exported the photos from DxO PhotoLab 6 and DxO PureRaw 2 by converting to a DNG file. With most photos, this caused no issues, but the Andean Cock-of-the-rock was a different case. Its deep orange color is a tough test for cameras and editing programs alike.

For point of reference, here’s how the orange color should look, as shown prior to exporting the image:

Export dialog Deep Prime XD

But here’s how it actually looked upon export to a DNG:

cock-of-the-rock_Color shift
Compare the color shift with the correctly converted crops above; NIKON D500 + 400mm f/2.8 @ 400mm, ISO 4500, 1/320, f/3.2

It almost looks pink, like a tamer version of a flamingo! At first I thought this was a problem with the XD algorithm, but upon exporting it with the original DeepPRIME algorithm as well, the problem persisted.

Only when I exported as a TIFF file instead of DNG did the color look normal again. I don’t know why that is exactly, but it was something I wanted to note in case you are noticing similar issues with the colors in your DNG files.

Summary

DxO has taken a step forward with its new way of pre-editing RAW files. DeepPRIME XDdelivers what its name promises. It actually manages to extract a slightly higher level of detail from files shot at high ISO than its predecessor DeepPRIME.

However, there are a few minor issues. The first problem I’ve encountered is that at very high ISOs, the new algorithm can create artifacts in areas that are out of focus. The second finding is color shift when converting to DNG, although this isn’t necessarily worse than with the previous DeepPRIME algorithm.

Summing up all the positives and negatives of the new DeepPRIME XD, the positive feelings prevail for me. If you frequently shoot at high ISOs and routinely face heavy noise, getting some software with the XD noise reduction algorithm could be useful for your work.

That said, at the moment, it’s only in the DxO PhotoLab 6 Elite software, which is fairly expensive at €219 (note that the cheaper ESSENTIAL Edition does not include the DeepPRIME XD). I hope that DxO adds this feature to their PureRaw software or PhotoLab Essential, because although it’s a useful improvement, it may not be worth the price if you wouldn’t use the other features of PhotoLab Elite.

Products You May Like

Articles You May Like

My Favorite Photography Deals of the Holiday Weekend
How to Use Healing Modes in Lightroom Classic
Footballer wipes hands on photographer’s bib in middle of World Cup USA v UK match
Balenciaga under fire for posing kids with BDSM props; blames it on the photographer
Square crop compositions – They’re not just for Instagram

Leave a Reply

Your email address will not be published. Required fields are marked *