Pages: [1] 2   Go Down

Author Topic: DxO PhotoLab 4 DeepPrime noise reduction  (Read 3335 times)

Cem

  • Full Member
  • ***
  • Offline Offline
  • Posts: 124
    • Photographs
DxO PhotoLab 4 DeepPrime noise reduction
« on: October 22, 2020, 11:01:29 am »

Yesterday, DxO has announced the latest version of PL, which is PL4. This has a so-called DeepPrime noise reduction mode which works (according to DxO) directly on the not yet demosaiced raw file and does the noise reduction and demosaicing in one go. I have been testing it using a range of my own high ISO images from various cameras and the results are astonishing. I have compared the PL4 noise reduction to LR/ACR/PS, Topaz Denoise AI and C1 Pro. PL4 was the best performer. If you shoot lots of high ISO/high noise images, I recommend that you download the trial version (30 days functional) and give it a go.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 18223
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #1 on: October 22, 2020, 12:14:40 pm »

Will it do this on NR on DNGs?
Logged
Author “Color Management for Photographers" & "Photoshop CC Color Management" on pluralsight.com

Cem

  • Full Member
  • ***
  • Offline Offline
  • Posts: 124
    • Photographs
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #2 on: October 22, 2020, 01:18:46 pm »

Hi Andrew,

As long as the camera body/sensor is supported, DNG file format works. I have tested with native dng files from a Pentax 645 as well as with raw files converted to dng using the adobe tools (both with embedding of the original raw and without).
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 18223
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #3 on: October 22, 2020, 01:27:23 pm »

Downloaded, indeed it does support my converted DNGs.
What I can't yet figure out is the main preview vs. the little one for DeepPrime and other NR. They don't match as seen below. Yet I can toggle corrections on and off and do see an improvement. When I export the raw, the TIFF looks like the little preview. So can't the 'main' preview show the results?
Also below, comparison of one very high ISO image in Dx0 vs. LR. I tried to get the overall tone of the two renderings as close as possible.
Logged
Author “Color Management for Photographers" & "Photoshop CC Color Management" on pluralsight.com

Cem

  • Full Member
  • ***
  • Offline Offline
  • Posts: 124
    • Photographs
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #4 on: October 22, 2020, 01:32:32 pm »

It is indeed a shortcoming that it does not display the fully DeepPrime rendered results on the main window. I think that it was a conscious decision since the processing may take a long time. One can see the full results only if the image is exported to disk and then re-opened in any image processing software.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 18223
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #5 on: October 22, 2020, 01:40:26 pm »

It is indeed a shortcoming that it does not display the fully DeepPrime rendered results on the main window. I think that it was a conscious decision since the processing may take a long time. One can see the full results only if the image is exported to disk and then re-opened in any image processing software.
Thanks for confirming that. Drag. Pretty fast updating and exporting with my GPU; should be an option to be on or off.
Logged
Author “Color Management for Photographers" & "Photoshop CC Color Management" on pluralsight.com

RT

  • Newbie
  • *
  • Offline Offline
  • Posts: 10
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #6 on: October 22, 2020, 06:42:20 pm »

Cem, I'm wondering if there is a significant difference from the current PhootLab 3?  That is, do you now/think it is worth the upgrade?

Rene
Logged

leuallen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 426
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #7 on: October 22, 2020, 11:53:47 pm »

I compared Dxo with Denoise. DXo is on par with Denoise except for dark under exposed shadows. Denoise is noticeably better there. M43 at ISO6400.

Do not like the preview. You have to have faith that it will do the job.
Logged

Cem

  • Full Member
  • ***
  • Offline Offline
  • Posts: 124
    • Photographs
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #8 on: October 23, 2020, 06:03:15 pm »

Cem, I'm wondering if there is a significant difference from the current PhootLab 3?  That is, do you now/think it is worth the upgrade?

Rene
Hi Rene,

I am a PL3 user and the first thing I did was to look for an answer to that question. The PL4 is definitely better than PL3 and it is worth the update imo.
Logged

RT

  • Newbie
  • *
  • Offline Offline
  • Posts: 10
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #9 on: October 23, 2020, 07:02:26 pm »

Thanks, CEM.  I'm on a little bit of a hiatus from intensive computer work due to recurring repetitive strain syndrome, but I'll probably upgrade just catch the discount.

Rene
Logged

IanSeward

  • Full Member
  • ***
  • Offline Offline
  • Posts: 175
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #10 on: October 30, 2020, 06:48:51 am »

It is indeed a shortcoming that it does not display the fully DeepPrime rendered results on the main window. I think that it was a conscious decision since the processing may take a long time. One can see the full results only if the image is exported to disk and then re-opened in any image processing software.

Same situation with the original PRIME. Works on bayered, rather than demosaiced data so very processor intensive. Other drawback that this causes is that it obviously only works on raw (Bayer) files. However, they do provide a "normal" denoise routine (HQ noise reduction) for rgb images. I wonder if they could develop the "AI" part of the denoise to work on rgb images in the future, similar to Topaz Denoise AI?

DXO do have the benefit of years of data files (100,000's) for camera/lenses measured in their optical labs measured at different ISO and aperture, a unique situation that they are now exploiting.

On the DXO forums the engineers have said that they are looking at trying to make the full preview available now that they are using the increased computer power offered by using the GPU (Nvidia 1060 or equiv recommended).

If anyone is building a new PC they would be wise to factor in this software trend to use the GPU :-)
Logged

RT

  • Newbie
  • *
  • Offline Offline
  • Posts: 10
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #11 on: November 04, 2020, 05:50:50 pm »

Hi,

There a new review of Photolab 4 at "Life After Photoshop"

https://lifeafterphotoshop.com/dxo-photolab-4-review/

Rene
Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13866
    • http://www.flickr.com/photos/bernardlanguillier/sets/
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #12 on: November 05, 2020, 07:00:38 pm »

The noise reduction with DeepPrime is pretty remarkable I find.

As we speak, my personal ranking for raw converters remains: C1 Pro, DxO, Lightroom.

But with PL4 DxO is getting real close to C1 Pro.

Cheers,
Bernard

Rhossydd

  • Sr. Member
  • ****
  • Online Online
  • Posts: 3142
    • http://www.paulholman.com
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #13 on: November 08, 2020, 05:38:15 am »

It's an interesting convertor certainly. I'm very impressed by the quality of the noise reduction and the overall quality of the end results, but why is it so very, very SLOW. Outputting files can be glacial. Maybe useful for the odd special occasion, but FORTY times slower than Capture One ?????

Time for exporting a Canon 5Div file here (Win 10 3.9ghz i7, 32gb ram, SSD drives, nvidea GTX750 gpu)
Capture One v20....... 4.7s
LR 6........................ 7s
Affinity*.................. 12s
DPP 4..................... 18s
DXO PL4...................3m 27s

<edit...... later after mote tests>
It looks like the huge times are down to the DeepPrime processing.
HQ -10s
Prime - 58s
DeepPrime - 3m 30s

The times above are with 'auto' performance selected as a preference. Switching to CPU only reduces time by about 10s on DeepPrime.
Force it to use the gpu (even though not fully supported) and it drops to 57s so a massive improvement, if still much slower than the competition.


*Affinity time is a little different as exporting is a two stage process, once to 'develop' the file then export to file format. 12s is time for both added together.
« Last Edit: November 08, 2020, 08:11:07 am by Rhossydd »
Logged

IanSeward

  • Full Member
  • ***
  • Offline Offline
  • Posts: 175
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #14 on: November 09, 2020, 06:06:33 am »

It's an interesting convertor certainly. I'm very impressed by the quality of the noise reduction and the overall quality of the end results, but why is it so very, very SLOW. Outputting files can be glacial. Maybe useful for the odd special occasion, but FORTY times slower than Capture One ?????

Time for exporting a Canon 5Div file here (Win 10 3.9ghz i7, 32gb ram, SSD drives, nvidea GTX750 gpu)
Capture One v20....... 4.7s
LR 6........................ 7s
Affinity*.................. 12s
DPP 4..................... 18s
DXO PL4...................3m 27s

<edit...... later after mote tests>
It looks like the huge times are down to the DeepPrime processing.
HQ -10s
Prime - 58s
DeepPrime - 3m 30s

The times above are with 'auto' performance selected as a preference. Switching to CPU only reduces time by about 10s on DeepPrime.
Force it to use the gpu (even though not fully supported) and it drops to 57s so a massive improvement, if still much slower than the competition.


*Affinity time is a little different as exporting is a two stage process, once to 'develop' the file then export to file format. 12s is time for both added together.

Increasingly modern software is utilising the GPU to boost performance. DeepPrime is doing demosaicing and noise reduction at the same time using code designed to use the GPU. Systems optimised for Adobe software, that has traditionally not made use of the GPU are at a disadvantage if used with other software. Up side is that a GPU upgrade, if desired is relatively easy and a second-hand "good" GPU can be used if needed.

I have a Nvidia 2070 and can process a Canon 5Div in 13 sec, less if batch processing.
Logged

Cem

  • Full Member
  • ***
  • Offline Offline
  • Posts: 124
    • Photographs
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #15 on: November 09, 2020, 08:08:42 am »

I just did some testing now re. the times:

CPU: Intel 9700K, running at 5000 Mhz
RAM: 32 GB
GPU: Nvidia 2070GTX Super

A7 Mk III raw files

GPU Only
HQ = 8  s
Prime = 38 s
DeepPrime = 18 s

CPU Only
HQ = 8  s
Prime = 38 s
DeepPrime = 70 s

As you can see, HQ and Prime modes do not make use of the GPU. But DeepPrime does, and the resulting 18s is 20 full seconds faster than the Prime method. GPU causes a 400% speed increase. Therefore, I have no valid reasons to use Prime, I can either use HQ for low ISO images without much noise or I can use DeepPrime for noisy images.

18 seconds is as far as I am concerned more than acceptable. No complaints on my part. It may be twice the time it takes for LR or C1, but it is still negligible since I process raw files one at a time and do not batch process hundreds of them in one go.

Logged

IanSeward

  • Full Member
  • ***
  • Offline Offline
  • Posts: 175
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #16 on: November 10, 2020, 01:12:47 pm »

I just did some testing now re. the times:

CPU: Intel 9700K, running at 5000 Mhz
RAM: 32 GB
GPU: Nvidia 2070GTX Super

A7 Mk III raw files

GPU Only
HQ = 8  s
Prime = 38 s
DeepPrime = 18 s

CPU Only
HQ = 8  s
Prime = 38 s
DeepPrime = 70 s

As you can see, HQ and Prime modes do not make use of the GPU. But DeepPrime does, and the resulting 18s is 20 full seconds faster than the Prime method. GPU causes a 400% speed increase. Therefore, I have no valid reasons to use Prime, I can either use HQ for low ISO images without much noise or I can use DeepPrime for noisy images.

18 seconds is as far as I am concerned more than acceptable. No complaints on my part. It may be twice the time it takes for LR or C1, but it is still negligible since I process raw files one at a time and do not batch process hundreds of them in one go.

It is worth remembering that DeepPrime is doing demosaicing and noise reduction. It is worth checking to see if you obtain improved sharpness at all ISO's via DeepPrime as some have reported.
Logged

Rhossydd

  • Sr. Member
  • ****
  • Online Online
  • Posts: 3142
    • http://www.paulholman.com
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #17 on: November 10, 2020, 06:03:01 pm »

It is worth remembering that DeepPrime is doing demosaicing and noise reduction.
So is every other raw converter, just in a different order.

I'm sure most people will understand that a better and more complex algorithm for noise reduction might take longer, but is that longer processing time worth waiting for ? that will depend on how important that noise reduction is to you and how many images you need to process.

It would also be good if the self configuration settings worked well, which they don't here.
Logged

IanSeward

  • Full Member
  • ***
  • Offline Offline
  • Posts: 175
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #18 on: November 11, 2020, 10:30:48 am »

So is every other raw converter, just in a different order.

I'm sure most people will understand that a better and more complex algorithm for noise reduction might take longer, but is that longer processing time worth waiting for ? that will depend on how important that noise reduction is to you and how many images you need to process.

It would also be good if the self configuration settings worked well, which they don't here.

Not everyone is getting the improved noise reduction results as DXO though :-)

It is always an issue when your software exceeds the capabilities of your hardware. :-)

There have been a number of threads over on an Amazon owned review site :-) It looks like if you specced your computer to only work with Adobe then the issue is people skimped on the GPU. No problem if you only use Adobe software. With CPU power having plateaued a few years ago software companies are harnessing the power of the GPU. Topaz is a prime example of this. In some ways it can be considered a bonus. I could get equivalent process times for a D850 file using Topaz Sharpen AI in Stabilise mode using a 9 year old i3820 CPU as a guy using a 24 core AMD Threadripper CPU; we both had the same Nvidia 1060 6Gb GPU :-) So upgrading my GPU had turned out to be a good investment.

Using the GPU is going to only increase going forward as it represents the best opportunity to increase the computing power available. In the case of DXO anyone with a "decent" GP, DXO recommend a Nvidia 1060 or equivalent as their minimum have found DeepPrime not only gives them superior but also faster results than PRIME. Win-Win if you have a GPU which meets the minimum spec.
Logged

Rhossydd

  • Sr. Member
  • ****
  • Online Online
  • Posts: 3142
    • http://www.paulholman.com
Re: DxO PhotoLab 4 DeepPrime noise reduction
« Reply #19 on: November 11, 2020, 11:24:08 am »

It looks like if you specced your computer to only work with Adobe then the issue is people skimped on the GPU.
You might bare in mind that Adobe were one of the first imaging companies to use GPU acceleration. When the Mercury playback engine was first released in Premiere Pro it made most other software video renderers look painfully slow.

I'm primarily pointing out that the auto configuration settings isn't working on PL4 to get the best from the software.

Saving a half a minute or so on rendering an image isn't worth having to spend £200 on a new graphics card. Everything else runs very sweetly on my system.

Logged
Pages: [1] 2   Go Up