Blog

Note:
In the mean­time, DxO has released ver­sion 6 of its RAW con­vert­er DxO Pho­to­Lab*. This arti­cle still refers to the pre­vi­ous ver­sion 5. The results when denois­ing with Deep­PRIME are prac­ti­cal­ly iden­ti­cal in all ver­sions, so every­thing pre­sent­ed here con­tin­ues to apply to it. In ver­sion 6, DxO has improved denois­ing even fur­ther and intro­duced the even more refined AI-based Deep­PRIME XP algo­rithm. More details can be found in my updat­ed review.

As I wrote in my pre­vi­ous arti­cle on DxO Pho­to­Lab 5, I’ve been using DxO Pho­to­Lab since ver­sion 4 to denoise my high-ISO shots using the awe­some Deep­PRIME process imple­ment­ed there, and exclu­sive­ly for that pur­pose. I con­tin­ue to do all fur­ther pro­cess­ing and all pro­cess­ing of all images that do not require this elab­o­rate denois­ing in Adobe Light­room. Since Ver­sion 1 of Light­room, I have also orga­nized and key­word­ed all my images there.

DeepPRIME

In ver­sion 4 of Pho­to­Lab, DxO intro­duced Deep­PRIME*, an arti­fi­cial intel­li­gence-based method for reduc­ing noise in pho­tos. I down­loaded the tri­al ver­sion out of curios­i­ty and was so thrilled with the results that I bought DxO Pho­to­Lab 4 just for denois­ing my high-ISO shots. A detailed report on my results can be found here. Since then, I’ve been using DxO Deep­PRIME reg­u­lar­ly and hap­pi­ly. With my Canon EOS R5, I now have no wor­ries about using ISO val­ues up to 12,800 when necessary.

Now DxO Pho­to­Lab has been updat­ed to ver­sion num­ber 5. Although many func­tions for RAW pro­cess­ing have been added and exist­ing ones improved, this is not rel­e­vant for me, since I only use the Deep­PRIME algo­rithm in Pho­to­Lab. Though you could also use the slimmed-down ver­sion DxO Pur­eRAW for this pur­pose, which I test­ed here, Pur­eRAW offers only a few set­ting options. It also sharp­ens the image too much for my taste and the inte­gra­tion into my usu­al work­flow with Light­room is not opti­mal either.

So, I down­loaded the tri­al ver­sion of Pho­to­Lab 5* first. This can be used for 31 days with­out any restric­tions. I find it very praise­wor­thy that DxO allows this long tri­al peri­od. This allows you to test all the func­tions exten­sive­ly and get a good idea of the program’s capa­bil­i­ties before you spend a rel­a­tive­ly large amount of mon­ey on it. For the ELITE Edi­tion - and only this one includes the Deep­PRIME algo­rithm - you have to pay 219,-€ (cur­rent­ly there is an intro­duc­to­ry dis­count of 30%*). The update from ver­sion 4 costs 79.99€.

What’s new in version 5?

I now want­ed to see which advan­tages are offered by ver­sion 5 to me. DxO states that the Deep­PRIME algo­rithm has become faster on PCs. This is what I checked first. Again, I used the test image of an Ibis (tak­en with my EOS R5 and ISO 12,800) used in my com­par­i­son arti­cle of Deep­PRIME with Topaz Denoise AI. If you want to try it your­self, you can still down­load the CR3 raw file here.

I first processed the ini­tial­ly rel­a­tive­ly dark and low-con­trast image in Adobe Light­room Clas­sic 10.4 accord­ing to my taste. The result can be found below:

With­out denois­ing in Light­room, the 100% crop looks like this:

First, I test­ed the speed of Deep­PRIME on my desk­top com­put­er. This is a 16-core AMD Thread­rip­per 2950X with 32GB of main mem­o­ry and a NVIDIA GeForce RTX 2080Ti. The RAW file was trans­ferred via the Pho­to­Lab Light­room plu­g­in. I processed the image mul­ti­ple times using Pho­to­Lab 4 and 5 respec­tive­ly. I set the lumi­nance in Deep­PRIME to 100. After­wards I processed the denoised ver­sion with the option

“Als DNG exportieren (Nur Rausch­min­derung und optis­che Korrekturen)”

and export­ed it back to Light­room. There­fore, the pro­cess­ing time of the 45 megapix­el file of the EOS R5 file with the Deep­PRIME algo­rithm was 26 sec­onds for both pro­gram ver­sions. I could not detect any accel­er­a­tion in ver­sion 5 on my desk­top computer.

After that, I used my new lap­top. It has an 8-core Intel Core i9-11980HK, 64GB main mem­o­ry and an NVIDIA GeForce RTX 3080 with 8GB of mem­o­ry. The per­for­mance in this case was aston­ish­ing: Pho­to­Lab 4 now only need­ed 16 sec­onds, Pho­to­Lab 5 was even fin­ished after 10 sec­onds! This real­ly con­firms DxO’s per­for­mance promise.

Although the two DNG files cre­at­ed in this way have slight­ly dif­fer­ent sizes, so they are not com­plete­ly iden­ti­cal, I could not detect any sig­nif­i­cant dif­fer­ences in qual­i­ty between the two ver­sions even when I looked close­ly at them in 100% mag­ni­fi­ca­tion. There­fore, here is only a 100% crop of the result with DxO Pho­to­Lab 5:

So far, so good and already known. The per­for­mance improve­ment on my cur­rent lap­top is quite nice, but does not yet jus­ti­fy the pur­chase of the new ver­sion for me. But there was still some­thing new in ver­sion 5:

Support for DJI Air 2S and Mini 2

DxO Pho­to­Lab 5 now sup­ports my two new­ly acquired drones, the DJI Mini 2 and the Air 2S, in addi­tion to many old­er DJI drones:

Espe­cial­ly the small sen­sors of the drones nat­u­ral­ly tend to pro­duce noise even at mod­er­ate ISO val­ues. That’s why I only use my DJI Air 2S with its native ISO val­ue of 100 when­ev­er pos­si­ble. Even at ISO 400, noise becomes notice­able and annoy­ing. I already showed an exam­ple of this in my first review of the DJI Air 2S:

Here again a detail in 100% mag­ni­fi­ca­tion, which demon­strates the already clear­ly vis­i­ble noise:

I was extreme­ly curi­ous to see what DxO Pho­to­Lab 5 could do with this image. After pro­cess­ing with Deep­PRIME, the result with the same set­tings in Light­room as for the image above looks like this:

And here is anoth­er 100% crop from the image above:

That looks pret­ty good already, does­n’t it? So I searched my Light­room library for anoth­er image of the DJI Air 2S with even more pro­nounced noise to test fur­ther. In the process, I found a shot of a lone tree at dusk that I had tak­en a few weeks ago in War­burg with the DJI Air 2S. The image was com­posed of 9 sin­gle shots with the panora­ma mode of the drone. The sin­gle shots were each tak­en in the Air 2S’s auto­mat­ic mode with f/2.8, 1/5s expo­sure time and ISO val­ues from 550 to 1170. Here at first my exten­sive­ly processed image in Lightroom:

The pic­ture is quite nice, but in the 100% view with­out fur­ther noise reduc­tion, which can be seen here, the noise is already very noticeable:

How­ev­er, when I effec­tive­ly de-noise it in Light­room, most of the details are lost, as you can see here:

I have there­fore denoised the sin­gle RAW files in DNG for­mat of the DJI Air 2S with Deep­PRIME and cre­at­ed a new panora­ma with Light­room from the 9 result­ing files. After fur­ther pro­cess­ing in Light­room it now looks like this:

And here again, the same sec­tion of the image denoised with Deep­PRIME in 100% view. You can clear­ly see that DxO Pho­to­Lab 5 pre­serves con­sid­er­ably more detail with the same lev­el of noise reduction:

The result is a clear improve­ment, espe­cial­ly con­sid­er­ing the adverse con­di­tions of the shot (long expo­sure time, small sen­sor, high ISO val­ue). Unfor­tu­nate­ly, how­ev­er, the panora­ma of the DxO DNG files assem­bled in Light­room shows sig­nif­i­cant vignetting. As a result, the lim­i­ta­tions of the com­pos­ite indi­vid­ual shots are also vis­i­ble upon close inspec­tion. The whole thing works even bet­ter if you stitch the panora­ma in the spe­cial­ized PTGui pro­gram. There you can also adjust the per­spec­tive. The final result of my new work­flow is shown here:

DJI Air 2S DNG -> DXO Pho­to­Lab 5 Deep­PRIME -> PTGui -> Lightroom

sieht dann so aus:

And Topaz Denoise AI?

As promised before, I also used the lat­est ver­sion (3.3.3) of Topaz Denoise AI on the panora­ma. Since DNG for­mats are still not sup­port­ed by Topaz Denoise AI, I used the already in Light­room stitched panora­ma and denoised it with Denoise AI using the auto­mat­ic set­ting “Severe Noise”. Denoise AI chose the parameters:

• Remove Noise 44
• Enhance Sharp­ness 23

The result­ing image then looks like this:

And for direct com­par­i­son again the 100% view:

While this is bet­ter than what I can achieve in Light­room alone. It shows much more detail, but still does­n’t come close to the result of Pho­to­Lab 5.

Resumée

All in all, DxO Pho­to­Lab ver­sion 5 prob­a­bly brings a lot of new fea­tures in image pro­cess­ing, but I don’t use them. The same applies to me for the now includ­ed sup­port for the Fuji X-Trans sen­sors. I could only repro­duce the stat­ed speed increase of the Deep­PRIME algo­rithm on my lap­top with cur­rent hard­ware. On my desk­top com­put­er, both ver­sions are equal­ly fast. The qual­i­ty of denois­ing with the AI-based Deep­PRIME algo­rithm is as good as with the pre­vi­ous ver­sion 4 and much bet­ter than I can man­age with Light­room alone or with Topaz Denoise AI. So far, an update would­n’t actu­al­ly be nec­es­sary for my work­flow - if I weren’t now using the new DJI drones in addi­tion to my Canon EOS bod­ies. Since these are only sup­port­ed by the new ver­sion, I there­fore pur­chased the update to ver­sion 5 of Pho­to­Lab in the end.

If DxO is used as a stand­alone RAW devel­op­er or for Fuji pho­tog­ra­phers with X-Trans sen­sors, the update may be worth­while even as it is, but that is up to each per­son to judge for them­selves. Thanks to the option to test DxO Pho­to­Lab with­out restric­tions for 31 days, this can also be exten­sive­ly checked before buying.

*= Affil­i­ate link