The purpose of DeepPrime 3 (Pure Raw 5 vs Pure Raw 4)

There have been several topics that have adversely compared PR5 with PR4. But I think that some users may be misunderstanding the nature of the changes implemented in PR5.

PR5 keeps the XD2/XD demosaicing/denoising software that was introduced in PR4. As far as I can tell, the results from this process in PR5 are identical with the results produced by the same process in PR4.

PR5 also introduces a new demosaicing/denoising routine: DeepPrime 3. This is not a better denoising routine than XD2 and is not described as such by DXO.

As stated in the user guide, DeepPrime 3 produces ā€œquality resultsā€ but not the same level of fine detail as XD2. The advantage of DeepPrime 3 is that it is much faster than XD2.

On my (Intel) iMac, DP3 processes a 70mb Leica Q3 file in under 20 secs, while XD processes the same file in about 40 secs.

To quote the User Guide:
ā€œDeepPRIME 3: This method uses AI and deep learning technologies, and produces high-quality results, but faster than the other two methods.
DeepPRIME XD2s/XD2: an advanced version of DeepPRIME that offers finer detail extraction. DeepPRIME XD2s goes even further with more details and no artifacts.ā€

Is this faster but not quite as detailed option worth the upgrade price? I’m not convinced it is for me. But, as always, YMMV.

Best wishes,

Peter

7 Likes

I’m still a bit puzzled that some people are finding noticeable differences between PureRAW 4’s and PureRAW 5’s DeepPRIME XD2S / XD.

I also see a little ā€œtensionā€ between denoising and detail enhancement.

One thing I would wish is that when making direct comparisons between PureRAW 4 and 5, people would use the DXO downloadable example files so we’re all on the same page.

1 Like

I’ve changed my mind about XD2s/XD being the same in PR4 and PR5. I now think I see a subtle difference in favour of the PR5 version. It is less contrasty and creates fewer artefacts where surfaces are (probably) textured but the texture is not well-resolved (e.g. because of sensor or focal length limitations).

In this screenshot (LR) the LHS shows a section of buildings from an image treated in PR4. The RHS shows the same section from the same image treated in PR5. I see fewer artefacts in the PR5 version (e.g. on the brick tower).

The wooden tower below the brick one seems better resolved by PR5, too.

3 Likes

Sorry to contradict you, but after multiple adjustment tests with files from different cameras (Nikon, Canon, Fuji, and Leica), I can confirm that the results obtained with version 5 are far below those obtained with version 4. This version 5 is of NO INTEREST, except that it is slightly faster.
I now only use version 4!
I regret pre-ordering; I’m still waiting for a response from DxO regarding a refund for my somewhat premature purchase!

2 Likes

Fair enough!

May I ask which of the options you chose in PR5? Was it XD2S/XD or Deep Prime 3? Or do you find that both options are inferior in some way to PR4?

I haven’t yet found the improvements in XD2S/XD are remarkable but, as I note in my second post above, I now think there are visible improvements when I process my OM-1 files.

Also, as I continue testing, DP3 seems to be consistently almost twice as fast to process files on my Intel IMac (although with lower quality de-noise just as DXO notes in the User Guide).

Peter

1 Like

I choose both and DeepPrime3 is worth than XD2S/XD. In allecase, PR5 is worth than PR4. It is blatant with an old CR2 file taken with an EOS 5D MKII .
The left part is with PR4, the right partin PR5, both with XD2S/XD

It has been demonstrated that there is a difference between PureRAW4 and PureRAW5 in the demosaicing.

It has also been suggested that final (output) sharpening be saved for your post-processing software (DXO, PhotoShop, LightRoom, etc.) rather than attempting too much on the input side.

1 Like

I’m willing to admit there’s a difference between PR4 and PR5, but the results are unacceptable. I feel like I’ve been ripped off by DxO for €80!
I just tried Topaz Photo AI this morning, and what a surprise! The results are remarkable, as long as you don’t push the sliders too far! I’ve just found the almost ideal software . So long DxO !

PR4 seems to be sharper

Here we go, direct comparison between ACR with details set to on. PR4 and PR5. No cropping or any other post-processing done. This is SOOC.

Here are 3 screenshots confirming the settings.

Shot with a Z8, 135mm Plena, ISO 64 f/4, 1/400 sec. The duck posed…

Screenshot 2025-04-21 202031.jpgScreenshot 2025-04-21 201849.jpgScreenshot 2025-04-21 202642.jpg

Here we have full-res images saved at the highest settings.

The first image is ACR, followed by PR4 and PR5

For high ISO images.

Here are older images in Fort Da Sota, Florida.

Z9 500pf (FW = Initial versions back in 3-2022)

Various shutter speeds.

On each set, the first is PR4, and the second is PR5, with the same setting, SOOC, and no processing.

ISO 4500

ISO 8000

ISO 16000

ISO 22800

ISO 25600

ISO 25600

Couple of comments:

  1. You are comparing Deep Prime 3 in PR5 to Deep Prime XDS in PR4. That is the wrong comparison. You need to compare Deep Prime XDS in both PR4 and PR5.

  2. The default setting in PR5 Deep Prime XDS are not equivalent to PR4 Deep Prime XDS default settings. DXO has made the defaults in PR5 so that less detail is extracted and turned down the default sharpening a bit. To do a comparison, you first need to find the best settings for each camera in PR5. PR5 has a lot more flexibility so you can cook the raw files into a DNG any way you want.

1 Like

The title of this discussion topic is, "The purpose of DeepPRIME 3 (Pure RAW 5 vs Pure RAW 4).

This can’t be stressed enough - it’s absolutely true. People who are comparing at default settings and saying ā€œPR5 is worthlessā€ aren’t comparing like for like. Yet, to be fair, DxO hasn’t made it clear that this is the intended behavior. I guess we have to read ā€œbetween the linesā€ when DxO says that PR5 now lets users create presets for different cameras and lets users vary the noise reduction and sharpening through masking.

3 Likes

This is exactly my point here.
The confusion is by design by DxO.

I have written about this on this forum, and complained to DxO directly.

There are 2 different types of processings going on.

  1. Lens correction profiles.
  2. Adding additional sharpening.

Soft = Lens correction profiles only.
Standard, Strong, Hard = Lens correction profile, PLUS good old-fashioned unsharp mask sharpening

Those 2 processes should not be married in one drop down.

The issue with Pure Raw performing image shapening before any editing is two fold: a) Sharpaning should be performed at a later stage in post processing. b) When using PureRaw sharpening, it can no longer be sharpened properly in Adobe ACR, LrC, etc, with additional sharpening because the image appears extremely unnatural over-sharpened, which is unacceptable.

With PR5, which setting will perform lens correction WITHOUT additional sharpening?

Thanks so much to Joel for taking the time to post so many comparisons at various ISOs. Even if it’s not apples-to-apples, as some say, I found the comparisons quite useful. For example, with the highest-ISO files, PR4’s XDS2 started outputting waxy mush similar to Topaz Photo AI. The noisier PR5 DP3 files kept more detail, less plastic (only with the highest ISOs). That said, I think I’ll stick with PR4 and be careful to dial back some of the crunchy detail you get with normal-ISO files.

I would love to hear from DxO what is the apples to apples comparison, it must be a way to just use lens correction without adding sharpnening. If there is no way, meaning DxO completely integrated lens correction with additional sharpening, it is bad news.

Regarding crazy high ISO; That’s not something I’m after in real life anyway. If I’m using PureRaw just for de-noising, I can continue to use Adobe De noise AI which is best in class. I want to be able to use DxO for its so called superior RAW conversation and their laboratory lens profiling.