Blog

Note:
In the mean­time, DxO has released ver­sion 8 of its new RAW con­vert­er, DxO Pho­to­Lab*. This arti­cle still refers to the pre­vi­ous ver­sion 4, in which arti­fi­cial intel­li­gence was intro­duced for the first time. What is pre­sent­ed here also applies to the cur­rent ver­sion; the results when denois­ing with Deep­PRIME are prac­ti­cal­ly iden­ti­cal in all ver­sions, so every­thing pre­sent­ed here con­tin­ues to apply. The few changes in ver­sion 5 can also be found in a fol­low-up arti­cle here. In ver­sion 6, DxO has improved denois­ing even fur­ther and intro­duced the new improved AI-based Deep­PRIME XD algo­rithm. More details about that can be found in my updat­ed review.

DxO has inte­grat­ed a new denois­ing tech­nol­o­gy, called Deep­Prime, in DxO Pho­to­Lab 4 for the first time, which is sup­posed to be able to denoise images much more effi­cient­ly, based on Deep learn­ing. The demo­saic­ing of the Bay­er matrix of the sen­sor has been ful­ly inte­grat­ed into the process.

I was now par­tic­u­lar­ly inter­est­ed to see what the high­ly acclaimed DxO noise reduc­tion method “Deep­PRIME” actu­al­ly does in practice.

How has it been up to now?

Although my new Canon EOS R5 already has a very pho­to­sen­si­tive sen­sor that pro­duces rel­a­tive­ly min­i­mal noise, in high­er ISO ranges you still have to denoise the images in post-pro­cess­ing reg­u­lar­ly. For years, I have been using the RAW for­mat for my cam­eras to have max­i­mum free­dom in edit­ing my images.

There­fore I have to “devel­op” all my pic­tures before I can use them. I have been using Light­room from Adobe, cur­rent­ly Light­room Clas­sic Ver­sion 10.0, for this pur­pose and for archiv­ing and tag­ging my images for many years. Over the gen­er­a­tions of Light­room, Adobe has con­tin­u­ous­ly improved its abil­i­ty to reduce the noise of high-ISO images.

In my post-pro­cess­ing with Light­room, I have made the expe­ri­ence with the EOS R5, that ISO val­ues of up to about 5000 can be used with­out sig­nif­i­cant com­pro­mis­es. Even up to ISO 12,800, the images are still usable, albeit with compromises.

An example

I spon­ta­neous­ly took the fol­low­ing pic­ture of our cat Tom a few days ago with the Sig­ma 60-600 at the EOS R5 with ISO 12800, reworked it in LR and espe­cial­ly denoised it accord­ing to my per­son­al taste. You always have to find an indi­vid­ual com­pro­mise between low noise and rich details. It’s a crop (por­trait from land­scape with 4098 x 5464 pix­els) and looks quite pass­able at first glance:

How­ev­er, if you look at the image at pix­el lev­el in 100%, you can already see the noise very clear­ly regard­less of the post-pro­cess­ing in Lightroom.

So I want­ed to see, if DxO Pho­to­Lab could do it even bet­ter with its new Deep­PRIME AI denois­ing tech­nol­o­gy. I down­loaded the tri­al ver­sion of DxO Pho­to­Lab 4. It can be used for 30 days with­out restric­tions. Dur­ing instal­la­tion, the pro­gram also auto­mat­i­cal­ly inte­grat­ed itself into Light­room and can be start­ed from there via the

File| Plug-in Extras | Trans­fer to DxO Pho­to­Lab 4

manu. I marked the raw image of our tom­cat in LR and passed it to DxO Pho­to­lab 4. There, I sim­ply select­ed the Detail tab in the right-hand con­trol pan­el and chose the new Deep­PRIME process in “DxO Denois­ing Technologies”.

I com­plete­ly omit­ted any fur­ther edit­ing, as I want to con­tin­ue edit­ing in Light­room. I then clicked the “Export to Light­room” but­ton at the bot­tom right and select­ed “Export as DNG (Denoise & Opti­cal Cor­rec­tions Only)” in the dia­log box that appeared:

The denois­ing takes about 30 sec­onds on my com­put­er (16 Core AMD Thread­rip­per 2950x, 32GB RAM and Geforce RTX 2080TI), then the image appears in Light­room as a dng-file and can be processed there. Is the result real­ly bet­ter than using Light­room alone?

And, yes, indeed, DxO’s Deep­PRIME actu­al­ly denois­es much bet­ter than I can man­age for myself inside Light­room. In the Light­room Clas­sic screen­shot above (100% zoom lev­el), you can see the Light­room sec­tion on the left (Noise Reduc­tion set­ting: Lumi­nance 60, Details 50, Con­trast 0) and the DxO Pho­to­Lab 4 sec­tion on the right. The dif­fer­ence is real­ly impres­sive. The noise in the back­ground is almost com­plete­ly elim­i­nat­ed and the image shows even more details than the Light­room ver­sion. Here you can see the whole image final­ly in the DxO version:

One more example

So I browsed through my archive and picked out an old­er image that showed mas­sive prob­lems with noise at that time: a pic­ture of the Milky Way in Kenya tak­en in 2015. At that time I took the pic­ture with the Canon TS-E 17mm f/4L on the Sony a7R at ISO 640 and had to bright­en it by about 4 f-stops in post-pro­cess­ing and cor­rect high­lights and depths mas­sive­ly. DxO Pho­to­Lab 4 can now han­dle this much bet­ter than before:

Here is a direct com­par­i­son with the noise reduc­tion in Light­room. On the left, noise reduc­tion in Light­room Clas­sic (lumi­nance 29) and on the right in DxO Pho­to­lab 4 with the default settings:

Over­all, the move to DxO Pho­to­Lab 4 shows a sig­nif­i­cant improve­ment in noise reduc­tion. So far, I would esti­mate the improve­ment at about 2 f-stops: An image processed in DxO Pho­to­Lab 4 at ISO 6400 would have a sim­i­lar noise behav­ior to an image processed in Light­room Clas­sic at ISO 1600. But I want­ed to inves­ti­gate this more systematically.

A third example

On our safari in Kenya at the turn of the year 2018 to 2019, we dis­cov­ered a Ser­val cat run­ning through the steppe on our last morn­ing gamedrive - just before the sun­rise. More infor­ma­tion and pic­tures of the beau­ti­ful cat can be found in my trav­el report.

How­ev­er, in the dark it was a real chal­lenge. With the 400mm f/2.8 at wide-open aper­ture and a crit­i­cal slow shut­ter speed of 1/250s on my 5DS R, I still had to raise the ISO val­ue to 6,400. Up to now, I was very sat­is­fied with the results after edit­ing in Light­room, but Pho­to­Lab 4 real­ly gets a lot more out of it, as the fol­low­ing two side-by-side images show:

A systematic investigation

I planned a series of expo­sures with ISO val­ues of 100 - 51200 from one and the same sub­ject and want­ed to com­pare the images using Light­room and DxO Pho­to­Lab 4 for noise reduc­tion. Since our cat did­n’t want to sit still, I had to use a stuffed mon­key as the subject:

Résumé

At all ISO lev­els, the image denoised by DxO Pho­to­Lab 4 looks much bet­ter than the ver­sion processed in Light­room alone. Of course, the noise reduc­tion in Light­room could have been “fine-tuned” a bit, but in DxO, as I said, I did­n’t change any set­tings at all, the auto­mat­ic adjust­ment works per­fect­ly. I am thrilled!

In direct com­par­i­son, I think that DxO, regard­ing the noise lev­el, is now even a good 3 f-stops bet­ter than I can man­age in Light­room alone, as the fol­low­ing two com­par­isons demonstrate:

I con­sid­er the detail and noise lev­els in the above com­par­isons to be quite com­pa­ra­ble, what do you think? Com­ments and sug­ges­tions for improve­ment would be high­ly welcome.

After exten­sive test­ing, I was com­plete­ly con­vinced by DxO’s noise reduc­tion, and the results are sen­sa­tion­al - bet­ter than any­thing I’ve been able to achieve in Light­room before. I’ve since reworked many of my old images tak­en at high ISO and was always impressed by the extra detail DxO Pho­to­Lab 4 was able to reveal.

How­ev­er, as a long-time Adobe Light­room user, I could not get used to fur­ther edit­ing my images with Pho­to­Lab 4. I miss the data­base mod­ule, the abil­i­ty to work with my add-ons and Pho­to­Lab 4 is also much slow­er than Light­room. Fur­ther­more, many cam­eras and lens­es are not sup­port­ed in Pho­to­Lab 4.

I there­fore use Pho­to­Lab 4 exclu­sive­ly for denois­ing prob­lem­at­ic high-ISO images, while I con­tin­ue to edit all oth­er images with Light­room Clas­sic. I export the prob­lem­at­ic images from Light­room to Pho­to­Lab 4, denoise them with Deep­PRIME and export the DNG back to Light­room with the option “Export as DNG (Noise reduc­tion & opti­cal cor­rec­tions only)”. 

The set­tings from Light­room that were changed before are applied to the DNG file, as long as they were pre­vi­ous­ly saved as XMP files. This is my default set­ting in Light­room. How­ev­er, the result of these set­tings in the DNG files is not always iden­ti­cal to the Light­room orig­i­nal. Fine cor­rec­tions are often nec­es­sary. The lat­est ver­sion of DxO Pho­to­Lab 4 now also turns sharp­en­ing and noise cor­rec­tion in Light­room off when the DNG file is trans­ferred back to Light­room, so that these cor­rec­tions are not per­formed twice.

Every­thing else (expo­sure and col­or cor­rec­tions, crop­ping, tag­ging, etc.) is done in my usu­al Light­room edit­ing steps. But even if I only use noise reduc­tion, the results have con­vinced me so much that I have bought DxO Pho­to­Lab 4 just for this pur­pose, although I don’t use the oth­er func­tions at all.

By the way, DxO has mean­while react­ed to the demand and now offers DxO Pur­eRAW, a stand­alone pro­gram that only pro­vides noise reduc­tion with Deep­PRIME and opti­cal lens cor­rec­tions, and can then pass the images processed in this way as DNG files to oth­er pro­grams for fur­ther edit­ing. A short review can be found here.

What­ev­er the case may be, DxO Pho­to­Lab is a sig­nif­i­cant step for­ward in terms of noise reduc­tion and is worth an absolute recommendation.