DxO DeepPrime XD Review – An Evolution of Noise Reduction

Synthetic intelligence, or quite machine studying, is now a everlasting half of many photograph enhancing applications. I not too long ago examined one such machine studying program for noise discount and sharpness, known as DxO PureRaw 2. Within the meantime, nevertheless, the French firm DxO reworked their de-noising module known as DeepPRIME (which is an element of PureRaw 2) and launched a brand new model known as DeepPRIME XD.

The suffix XD stands for “further element.” This provides us a touch of what to anticipate from the brand new algorithm. However how does it carry out? That’s what I needed to cowl in at this time’s fast evaluation.

To start out, I ought to say that I hoped to see DeepPRIME XD built-in into DxO PureRaw 2 through an replace. In truth, PureRaw 2 has develop into nearly an integral half of my enhancing workflow. Sadly, it didn’t occur. So perhaps in PureRaw 3? In any case, for now, it’s solely present in DxO PhotoLab 6.

As such, DxO PhotoLab 6 has taken up non permanent residence in my laptop. Solely time will inform if it should keep there completely. I’ve left all its enhancing and cataloguing options apart and centered my consideration solely on DeepPRIME XD.

Processing Pace

I chosen a set of eleven pictures for enhancing, taken principally with Nikon D500 and Nikon D750 cameras. The lenses used have been a Nikon 400mm f/2.8 G or a Nikon 200-500 f/5.6 zoom lens. I point out this reality as a result of each the camera and the lens should be in this system’s database so as to use its full potential. That’s why I didn’t embody the Nikon Z9 with the brand new telephoto lenses within the take a look at. Whereas the Nikon Z9 is acquainted to DxO, the Nikon 400 f/4.5 and 400 f/2.8 TC S lenses weren’t but within the database on the time of my testing.

I processed the whole sequence of images twice to attenuate any attainable error. The ensuing processing instances have been totally different for every technique, and never by a small quantity. DxO PhotoLab 6 processed the images the quickest, at 1 min. 12sec., after I set the older DeepPRIME algorithm. The identical algorithm, however in DxO PureRaw 2, accomplished the job in 1 min. 52 sec. And at last, my laptop’s processor had probably the most work to do with the brand new DeepPRIME XD. It took a protracted 3 min. 56 sec. – so, the brand new XD algorithm is hardly quick.

However what issues probably the most is picture high quality. Does it fare higher there?

Noise Reduction High quality

DxO claims that the usual DeepPRIME algorithm ought to cut back the noise degree by 2 stops, whereas DeepPRIME XD will present a distinction of 2.5 stops or extra. These are advertising numbers, of course, however they at the least present that DxO considers the XD model of the algorithm to be superior.

How does this distinction be mirrored within the images themselves? In my view, the distinction is noticeable instantly, should you zoom in sufficient.

Check out the 100% crops under. DeepPRIME XD (the picture on the suitable) clearly has the sting over the older algorithm on the left.

What ISO did I exploit for this photograph? A whopping ISO 20,000! I don’t shoot at ISO 20,000 day-after-day, however I can think about conditions the place it turns out to be useful, particularly in dense forests.

Though the DeepPRIME XD algorithm appears cleaner and fewer “blocky” at first look, word that it additionally created some artifacts underneath the fowl’s beak and behind its head. The older DeepPRIME algorithm could not look as clear, however the artifacts aren’t as apparent within the picture.

Preservation of Particulars

The standard selection in noise discount is fairly easy: Both you get a photograph with minimal noise and minimal element, or extra element however extra noise. If you need each, you usually want to make use of masks and apply noise discount solely the place it’s probably the most wanted, like out-of-focus areas.

Synthetic intelligence algorithms have improved this tradeoff, but it surely’s nonetheless an unavoidable half of images. That is the place I used to be most curious to see how DeepPRIME XD would examine to the earlier model.

The unique DxO DeepPRIME is on the left, and DxO DeepPRIME XD is on the suitable:

After I examine the outcomes, I’m not upset. You possibly can see that the sting sharpness has elevated barely within the XD model, and the quantity of noise within the out-of-focus background has decreased. Not like the prior take a look at, that is positively a win for the newer algorithm.

Right here’s one other instance under. Try the crest of the Andean Cock-of-the-rock. I believe the distinction is actually seen. (As earlier than, the older algorithm is on the left, and the newer XD is on the suitable.)

Shade Preservation

That is the place issues get a bit sophisticated. I exported the images from DxO PhotoLab 6 and DxO PureRaw 2 by changing to a DNG file. With most images, this prompted no points, however the Andean Cock-of-the-rock was a unique case. Its deep orange colour is a tricky take a look at for cameras and enhancing applications alike.

For level of reference, right here’s how the orange colour ought to look, as proven previous to exporting the picture:

However right here’s the way it truly appeared upon export to a DNG:

It nearly appears pink, like a tamer model of a flamingo! At first I assumed this was an issue with the XD algorithm, however upon exporting it with the unique DeepPRIME algorithm as properly, the issue endured.

Solely after I exported as a TIFF file as an alternative of DNG did the colour look regular once more. I don’t know why that’s precisely, but it surely was one thing I needed to notice in case you’re noticing comparable points with the colours in your DNG information.

Abstract

DxO has taken a step ahead with its new method of pre-editing RAW information. DeepPRIME XDdelivers what its title guarantees. It truly manages to extract a barely increased degree of element from information shot at excessive ISO than its predecessor DeepPRIME.

Nevertheless, there are a number of minor points. The primary drawback I’ve encountered is that at very excessive ISOs, the brand new algorithm can create artifacts in areas which are out of focus. The second discovering is colour shift when changing to DNG, though this isn’t essentially worse than with the earlier DeepPRIME algorithm.

Summing up all of the positives and negatives of the brand new DeepPRIME XD, the optimistic emotions prevail for me. In case you incessantly shoot at excessive ISOs and routinely face heavy noise, getting some software program with the XD noise discount algorithm might be helpful to your work.

That mentioned, in the meanwhile, it’s solely within the DxO PhotoLab 6 Elite software program, which is pretty costly at €219 (word that the cheaper ESSENTIAL Version doesn’t embody the DeepPRIME XD). I hope that DxO provides this function to their PureRaw software program or PhotoLab Important, as a result of though it’s a helpful enchancment, it might not be definitely worth the worth should you wouldn’t use the opposite options of PhotoLab Elite.